Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE
KASIMIR ALEXANDER ORLOWSKI June, 2019
SUPERVISORS:
Dr. D.B.P. Shrestha
Prof. Dr. V.G. Jetten
IMPACT ASSESSENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE
KASIMIR ALEXANDER ORLOWSKI Enschede, The Netherlands, June, 2019
Thesis submitted to the Faculty of Geo-Information Science and Earth
Observation of the University of Twente in partial fulfilment of the requirements
for the degree of Master of Science in Geo-information Science and Earth
Observation.
Specialization: Applied Earth Sciences, with specialization in Natural Hazards,
Risk and Engineering
SUPERVISORS:
Dr. D.B.P. Shrestha
Prof. Dr. V.G. Jetten
THESIS ASSESSMENT BOARD:
Prof. Dr. F.D. van der Meer (Chair)
Dr. C. Mannaerts (External Examiner, ITC – WRS Department)
DISCLAIMER
This document describes work undertaken as part of a programme of study at the Faculty of Geo-Information Science and
Earth Observation of the University of Twente. All views and opinions expressed therein remain the sole responsibility of the
author, and do not necessarily represent those of the Faculty.
i
ABSTRACT
Flash floods frequencies and magnitudes are increasing, influenced by climate change and land cover
alterations. To reduce losses by floods, hazard modelling is crucial. However, soil information which are
vital for modelling are often missing or of insufficient quality. Therefore, the publicly available global digital
soil database SOILGRIDS250m potentially represents a way to bridge the gap between data availability and
data demand. So far, its applicability for local scale hydrological modelling has not been sufficiently
investigated. In the light of that, this study focused on the analysis of two soil datasets, (I) detailed field data
(FD) and (II) SoilGrids (SG) in order to assess their similarity and to evaluate the sensitivity of flood
dynamics to their soil hydraulic properties (SHPs), and to different soil depths when applied in an integrated
flood model. Furthermore, the effects of land cover change on runoff generation and flood behaviour were
investigated. The first part of this research was dedicated to the soil data analysis. In the course of that, soil
properties of FD and SG were examined independently in relation to land cover and the terrain.
Subsequently, in a comparative assessment, the similarity between both datasets was quantified. The second
part of the study focused on land cover mapping with Google Earth Engine, and foremost on the
identification of land cover changes. In the third part, the preceding efforts were joined to build the input
for the integrated flood model openLISEM. The model was applied with: (I) SHPs derived in the laboratory
(FD); (II) SHPs predicted with pedotransfer functions (SG); and (III) with changing land cover information.
Results indicate that the FD and SG do not share many commonalities. FD is highly influenced by land
cover, whereas SG variability is limited throughout the watershed. Soil properties such as clay content, bulk
density and soil organic matter were overestimated by SG. Therefore, the use of SG led to far-reaching
consequences in the hydrology, including a considerable increase in flood extent, depth and duration.
Increasing soil depth influenced both datasets similarly by promoting infiltration and reducing surface water.
However, using FD, the flood dynamics were more sensitive to changes in soil depth. Changes in land cover
were predominately represented by deforestation and subsequent fruit tree cultivation. Changed land cover
information affected flood dynamics only minor, but an increase in runoff amounts was apparent. Quality
issues of the digital elevation model, including errors in elevation and flow connectivity, impeded the model
calibration efforts and led consequently to wrong flood patterns. However, as overall flood quantities are
expected to be correct, the conclusions for the objectives set were not invalidated. Future research should
continue exploring SG data applicability for hydrological modelling as it represents a valuable source of
information. To be able to make a profound statement about SG quality, it is necessary to conduct studies
in various regions of the world. This will help to investigate its quality dependency on factors such as terrain,
climate and vegetation.
Keywords: Hydrological modelling, openLISEM, Flash flood, SoilGrids, Soil properties, Land cover change,
Google Earth Engine, Thailand, Asia
ii
ACKNOWLEDGEMENTS
I would like to express my profound gratitude to all three supervisors who accompanied me during this
enriching journey, Dr. Dhruba Shrestha, Prof. Dr. Victor Jetten and Dr. Dinand Alkema. I have enjoyed
learning from and with you very much, whether being in Thailand in the jungle, or in your office at ITC.
Further, it is worth to note that I’m very grateful to Bastian Bout who lent me his brain from time to time,
and to my friends and companions Fernandes Fernandes Brenner and Felipe Augusto Fonseca Arévalo for
their support.
Ms. Nanette Kingma, you always told us: “you are in the driving seat”. However, sometimes it was good
that you helped us to navigate.
Furthermore, I express my sincere gratitude to my supervisors at the Asian Disaster Preparedness Center in
Bangkok, Dr. Peeranan Towashiraporn and Prof. Dr. Dr. Farrukh Chishtie, who facilitated my graduate
research within the SERVIR-Mekong project. Also, a big thanks goes to my dear colleagues and friends Dr.
Ate Poortinga and Mr. Biplov Bhandari, you two taught me a lot and increased my fascination to RS once
more, thank you!
Last but not least, I also would like to thank the staff of the Naresuan University – Phitsanulok, to provide
us with the necessary laboratory equipment and guidance.
Kasimir Alexander Orlowski
Enschede, June 2019
iii
LIST OF ABBREVIATION
ASF - Alaska Satellite Facility
BRDF - Bidirectional Reflectance Distribution Function
Db - Bulk Density
DEM - Digital Elevation Model
DEMm - DEM manipulated
DEMo - DEM original
DEMv - DEM without vegetation
EBBI - Enhanced Built-up and Bareness Index
ETM - Enhanced Thematic Mapper
EVI - Enhanced Vegetation Index
FAO - Food and Agriculture Organization of the United Nations
FD - Field Data
GEE - Google Earth Engine
GLAS - Geoscience Laser Altimeter System
IBI - Index-based Built-up Index
IDF - Intensity Duration Frequency
ISRIC International Soil Reference and Information Centre
Ks - Saturated Hydraulic Conductivity
LAI - Leaf Area Index
LCCS - Land Cover Classification System
LOI - Loss-on-Ignition
MNDWI - Modified Normalised Difference Water Index
NBR - Normalised Burn Ratio
ND - Normalised Differences
NDBI - Normalised Difference Built-up Index
NDVI - Normalised Difference Vegetation Index
NDWI - Normalised Difference Water Index
NIR - Near Infrared
iv
OLI - Operational Land Imager
OSM - OpenStreetMap
PCA - Principal Component Analysis
PTFs - Pedotransfer Functions
PSD - Particle Size Distribution
Psi - Matric Suction
RF - Random Forest
RLCMS - Regional Land Cover Monitoring System
RP - Return Period
RR - Random Roughness
SAVI - Soil-Adjusted Vegetation Index
SCS - Sun-Canopy-Sensor
SG - SoilGrids
SHP - Soil Hydraulic Properties
SL1 - Soil Layer 1
SL2 - Soil Layer 2
SOC - Soil Organic Carbon
SOM - Soil Organic Matter
SRTM - Shuttle Radar Topography Mission
SWIR - Shortwave Infrared
tcAngle - Tassel Cap Angle
tcDist - Tassel Cap Distance
TDOM - Temporal Dark Outlier Mask
TIRS - Thermal Infrared Sensor
USGS - U.S. Geological Service
WB-C - Walkley and Black Carbon
v
TABLE OF CONTENTS
1. INTRODUCTION ........................................................................................................................................... 1
1.1. Background .............................................................................................................................................. 1 1.2. Problem statement .................................................................................................................................. 4 1.3. Objectives and research questions ....................................................................................................... 5
2. RESEARCH AREAS ........................................................................................................................................ 6
3. METHODOLOGY .......................................................................................................................................... 8
3.1. Data ........................................................................................................................................................... 8 3.2. Soil data analysis ...................................................................................................................................... 8
3.2.1. Pedotransfer functions ...................................................................................................................................9 3.2.2. Independent soil assessment.........................................................................................................................9 3.2.3. Comparative soil assessment ..................................................................................................................... 10 3.2.4. Soil sampling strategy .................................................................................................................................. 10 3.2.5. Soil sampling ................................................................................................................................................. 11 3.2.6. Laboratory analysis ...................................................................................................................................... 12 3.2.7. Slope unit delineation .................................................................................................................................. 12
3.3. Land cover mapping ............................................................................................................................ 13 3.3.1. Random forest classifier ............................................................................................................................. 14 3.3.2. Land cover typology .................................................................................................................................... 15 3.3.3. Image processing ......................................................................................................................................... 16 3.3.4. Covariate layers ............................................................................................................................................ 18 3.3.5. Reference data .............................................................................................................................................. 21 3.3.6. Machine learning .......................................................................................................................................... 21 3.3.7. Validation data .............................................................................................................................................. 22 3.3.8. Accuracy assessment ................................................................................................................................... 22 3.3.9. Land cover change analysis ........................................................................................................................ 23
3.4. OpenLisem flood model .................................................................................................................... 24 3.4.1. Data preparation .......................................................................................................................................... 25 3.4.2. Model calibration ......................................................................................................................................... 29
4. RESULTS AND DISCUSSIONS ............................................................................................................... 30
4.1. Soil analysis ........................................................................................................................................... 30 4.1.1. Slope units ..................................................................................................................................................... 30 4.1.2. Influence of land cover on soil properties .............................................................................................. 31 4.1.3. Field data related to the terrain ................................................................................................................. 35 4.1.4. SoilGrids variability in the landscape ....................................................................................................... 38 4.1.5. Similarity analysis of SoilGrids and field data ......................................................................................... 40 4.1.6. Summary of soil analysis ............................................................................................................................. 42
4.2. Land cover analysis .............................................................................................................................. 42 4.2.1. Land cover change analysis ........................................................................................................................ 45
4.3. Flash flood modelling ......................................................................................................................... 48 4.3.1. Final model parametrisation ...................................................................................................................... 48 4.3.2. Calibration and the effects of different elevation models .................................................................... 50 4.3.3. Sensitivity of flood dynamics to soil information .................................................................................. 52 4.3.4. Effects of land cover change on flood dynamics ................................................................................... 57
5. CONCLUSION AND RECOMMENDATIONS .................................................................................. 62
ANNEXES .................................................................................................................................................................. 74
vi
LIST OF FIGURES
Figure 1. Location map of Ban Da Na Kham and Laplae watershed. ......................................................................................6 Figure 2. Climate diagram of the research area. Based on data from the Thai Meteorological Department for
Uttaradit city station (ID: 351002) for the period 2006 to 2018. .....................................................................................7 Figure 3. Slope positions in a natural landscape; adapted from Schoeneberger et al. (2012). ........................................... 11 Figure 4. Common landforms recognisable with Geomorphon (Jasiewicz & Stepinski, 2013). ....................................... 12 Figure 5. Workflow of the land cover classification; adapted from Saah et al. (2019). ....................................................... 13 Figure 6. Random Forest classification; V = random variables and S = random samples. ............................................... 14 Figure 7. Land cover classes in the research areas. a) Cropland, b) Orchard, c) Teak Plantation, d) Mixed Forest, e)
Urban, f) Water Bodies (Google Earth Pro, 2019). ........................................................................................................ 16 Figure 8. Locations of historical flood height measurements in Laplae city (left) and flood mark painted on a power
pole (right). ............................................................................................................................................................................. 29 Figure 9. Slope position map. ........................................................................................................................................................ 30 Figure 10. Variability of soil properties per land cover; a) Saturated Hydraulic Conductivity, b) Soil Organic Matter,
c) Porosity, d) Bulk Density. .............................................................................................................................................. 32 Figure 11. Variability of soil properties per slope unit; a) clay, b) sand, c) silt, d) Saturated Hydraulic Conductivity, e)
Porosity, f) Bulk Density, g) Organic Matter Content. .................................................................................................. 37 Figure 12. Land cover map Ban Da Na Kham watershed, a) 2005 and b) 2018. ............................................................... 44 Figure 13. Change map of Ban Da Na Kham watershed. ....................................................................................................... 46 Figure 14. Main road in 2005 a) and 2018 b); Source Google Earth Pro. ........................................................................... 46 Figure 15. Design storm rainfall graph. ....................................................................................................................................... 48 Figure 16. Soil depth map for scenario D1. ................................................................................................................................ 49 Figure 17. Flood simulation results in Laplae watershed using different DEMs: a) DEMo; b) DEMv; c) DEMm and
d) SRTM DEM. .................................................................................................................................................................... 51 Figure 18. Example of DEM irregularities in cropland areas; a) section of land cover map, b) flood simulation with
deep water ponds in parts of the cropland, c) hillshade showing undulating surface of cropland area, d) photo taken with view in north-east direction standing at black dot c). ................................................................................. 52
Figure 19. Maximum flood depth (m) for field data (FD) and SoilGrids (SG) for different soil depths: a) FD-D1, b) FD-D2, c) FD-D3, d) FD-D4, e) SG-D1, f) SG-D2, g) SG-D3 and h) SG-D4, D1= min: 0.36 m; max: 2.04 m, D2= min: 1.36 m; max: 3.04 m, D3= min: 2.36 m; max: 4.04 m, D4= min: 3.36 m; max: 5.04 m. ................ 54
Figure 20. Maximum flood duration (hr) using field data a) and SoilGrids b) with soil depth scenario D1 (min: 0.36 m; max: 2.04 m). Zoomed area shows how flood time changes when using different soil information. ............. 55
Figure 21. Total infiltration (mm) for field data a) and SoilGrids b) using soil depth scenario D1 = (min: 0.36 m; max: 2.04 m). Zoomed section of a) shows high infiltration values next to impermeable areas. Zoomed section of b) shows artificial border between low and high infiltration values. ...................................................................... 56
Figure 22. Maximum flood depth for 2005 a) and 2018 b). Zoomed in area indicates an increase in maximum flood depth in the lowland area due to land cover changes. .................................................................................................... 60
Figure 23. Flood duration for 2005 a) and 2018 b). Zoomed in area indicates prolonged floods in the lowland areas due to land cover changes. .................................................................................................................................................. 61
vii
LIST OF TABLES
Table 1. Data used and its source and properties. ........................................................................................................................8 Table 2. Soil samples per slope position and land cover. ......................................................................................................... 12 Table 3. Band combinations for normalised differences; adapted from Saah et al. (2019). ............................................... 19 Table 4. Number of training data points collected in each class. ........................................................................................... 21 Table 5. Number of field validation points for each class. ...................................................................................................... 22 Table 6. Validation data points for 2005 and 2018. .................................................................................................................. 22 Table 7. Main input data required for openLISEM. ................................................................................................................. 25 Table 8. Surface parameter for openLISEM. ............................................................................................................................. 27 Table 9. Daily rainfall prior, during (red square) and after the event. .................................................................................... 28 Table 10. Confusion matrix of slope unit classification. .......................................................................................................... 31 Table 11. Statistics of soil properties per land cover. .............................................................................................................. 31 Table 12. Correlation (r) of ancillary variables with Ks per land cover. ................................................................................. 33 Table 13. Soil properties per orchard subclass. .......................................................................................................................... 34 Table 14. Soil properties per slope unit. ...................................................................................................................................... 35 Table 15. Correlation of terrain derivates with soil properties. ............................................................................................... 36 Table 16. Correlation of Ks per slope unit with soil properties. ............................................................................................. 36 Table 17. Statistics of SoilGrids soil properties per land cover. ............................................................................................. 39 Table 18. Statistics of SoilGrids soil properties per slope unit................................................................................................ 39 Table 19. Cosine Similarity per Position. .................................................................................................................................... 40 Table 20. Cosine Similarity per Land Cover. .............................................................................................................................. 40 Table 21. Results of the test of normality. .................................................................................................................................. 41 Table 22. Output of Wilcoxon Signed-Rank test on SoilGrids derived properties. ............................................................ 41 Table 23. Land cover area for 2005 and 2018. ........................................................................................................................... 43 Table 24. Accuracy assessment of land cover map 2018. ........................................................................................................ 45 Table 25. Accuracy assessment of the land cover map 2005. .................................................................................................. 45 Table 26 Land cover change matrix of Ban Da Na Kham watershed (2005 to 2018)........................................................ 47 Table 27. Soil hydraulic properties SL1. ...................................................................................................................................... 49 Table 28. Soil hydraulic properties SL2 and texture class per slope position. ...................................................................... 49 Table 29. Water balance for eight model simulations using field data (FD) and SoilGrids (SG) data by using four soil
depth scenarios; (SG-FD) represents the difference between SG and FD for each soil depth scenario. ............. 53 Table 30. Flood depth related to soil data information and total reduction from D1 to D4. ........................................... 55 Table 31. Water balance of Ban Da Na Kham watershed for 2005 and 2018. .................................................................... 58 Table 32. Runoff amounts per land cover for 2005 and 2018. ............................................................................................... 59 Table 33. Flood depths and their corresponding extent due to land cover change. ........................................................... 60
viii
LIST OF ANNEXES
Annex 1. Geological map of Uttaradit (Ministry of Natural Resources and Environment Thailand, 2019) ................. 75
Annex 2. Pedotransfer script by Jetten and Shrestha (2018) based on the equations of Saxton et al. (2006) ............... 76
Annex 2.1. Soil sampling locations Ban Da Na Kham watershed ......................................................................................... 78
Annex 2.2. Morphological properties of the soil for each site (field measurements) ........................................................ 79
Annex 2.3. Physical and chemical properties of the soil for each site (field measurements) ........................................... 80
Annex 2.4. Description of laboratory measurements for soil property determination ..................................................... 81
Annex 2.5. Physical and chemical properties of SoilGrids for each site .............................................................................. 83
Annex 3. Description of Landsat images used for yearly and seasonal composites........................................................... 84
Annex 3.1. Google Earth Engine code for cloud removal by Saah et al. (2019) ................................................................ 86
Annex 3.2. Google Earth Engine code for cloud shadow removal by Saah et al. (2019) ................................................. 87
Annex 3.3. Google Earth Engine code for BRDF correction by Saah et al. (2019) .......................................................... 88
Annex 3.4. Google Earth Engine code for topographic correction by Saah et al. (2019) ................................................ 92
Annex 3.5. Google Earth Engine code to generate covariate layer (seasonal) adapted from Saah et al. (2019) ........... 95
Annex 3.6. Google Earth Engine code to generate covariate layer (yearly) adapted from Saah et al. (2019) ............... 99
Annex 3.7. Selected covariates layers for the classifications of 2005 and 2018 ................................................................ 103
Annex 4. IDF curves for Uttaradit province (Rittima et al., 2013) ..................................................................................... 104
Annex 5. Stream dimension measurements, locations and average values for validation ............................................... 105
Annex 5.1. Channel maps Ban Da Na Kham a) and Laplae b) watershed ........................................................................ 106
Annex 6. PCRaster script for soil depth map generation based on Kuriakose et al. (2009) ........................................... 107
Annex 7. Soil sampling site specific Cosine Similarity results .............................................................................................. 108
Annex 8. Land cover map Laplae watershed (2005) .............................................................................................................. 109
Annex 9. Gumble plot of daily long-term precipitation measurements (1951-2018). Red circle represents the flash
flood event in 2006. ............................................................................................................................................................ 110
Annex 10. Simulated flood height measurements in comparison to measured flood height points ............................. 111
Annex 11. Physical property layers of SoilGrids for soil layer 1 (5 cm) .............................................................................. 112
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
1
1. INTRODUCTION
1.1. Background
Hydrological hazards such as floods and droughts cause the loss of lives and economic damage around the
world. According to the World Disaster Report 2016, in the period between 2006 – 2015 flood hazards were
reported the deadliest, which also caused the highest economic loss as compared to the effects of other
natural hazards such as earthquakes and storms. With almost 700 events, Asia is the continent which was
affected most by flooding (IFRC, 2016). The 2011 flood was, for instance, one of the most disastrous events
in the recent history of Thailand, with a death toll of 884 and millions left homeless or displaced (Aon
Benfield, 2012). Another example is the year 2015 when Myanmar was devastated with flood problems.
Exceptional strong monsoon rains triggered landslides, flash floods, and river floods, with 69 casualties
affecting approximately 250 thousand people, countrywide. Moreover, more than 520 thousand acres of
agriculture land were destroyed (USAID, 2015).
Model-based projections and international long-term trend studies of hydrological processes prognoses an
increase of frequency and intensity of rainfall events in many parts of the world in the future due to the
effects of climate change (IPCC, 2014). Consequently, urgent attention should be drawn to flash floods to
prevent future disastrous events by means of capacity building and awareness raising activities.
Understanding the underlying processes and, favourable conditions for formation as well as potentially
triggering and influencing factors can provide implications for flood risk management and therefore prevent
the loss of human lives and economic assets.
Flash floods can be characterised by their temporal and spatial scale. Bout and Jetten (2017) associate them
with local rainfall events with high intensities and short durations, occurring mostly in mountainous
upstream watersheds. As such, they are related to short watershed response times with rapid increase and
release of discharge. Resulting floods may last several hours, but durations are rarely exceeding one day
(Bout & Jetten, 2017; Marchi et al., 2010). Therefore, it can be argued, that flash flood generation and
behaviour are among others related to the shape and size of the watershed (e.g., circular or elongated) and
to the pattern of the rainfall event (intensity and duration).
Runoff represents the main transport process for flood water during flash floods. Factors influencing runoff
generation are manifold. Rainfall represents the most fundamental factor. However, also hydrological pre-
conditions (e.g., initial soil moisture), soil physical properties, terrain (e.g., slope gradient), and land use and
land cover are decisive for runoff occurrence (Marchi et al., 2010). Physical properties comprise particle size
distribution (PSD), porosity, water retention properties and hydraulic conductivity of the soil. PSD provides
information about the grain size distribution of soil, hence about the percentage of sand, silt, and clay, which
affects soil hydraulic properties (SHPs) that are closely linked to runoff generation.
Decisive SHPs for runoff generation are infiltration, porosity and saturated hydraulic conductivity (Ks)
(Marchi et al., 2010). As described by Schaetzl and Anderson (2005), infiltration is the process of water
entering into the soil. Its rate depends, for example, on the pore size and initial soil moisture. Thereby,
porosity describes the pore space which can be filled by water, with increasing moisture content, the
infiltration rate decreases. Ks characterises the ease with which water or other fluids can move through the
soil. Fine textures (e.g., silt and clay) are smooth, having finer pores and a substantially greater volume of
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
2
open space compared to coarse textures like sandy soils. Therefore, their water holding and retention
capacity will be higher. At the same time, sandy soils have a higher Ks. Thus, the water can drain faster, and
less surface runoff will occur (Schaetzl et al., 2005).
Steady socio-economic developments can influence the occurrence of extreme flash flood events (Marchi
et al., 2010). Man-made land use and land cover changes like deforestation and urbanisation can modify the
hydrological processes of watersheds dramatically, and hence influence surface runoff characteristics and
flood dynamics (Sajikumar & Remya, 2015). In general, alterations can emerge for instance from compaction
(e.g., tillage), surface sealing (e.g., urban structures) and vegetation cover changes (e.g., deforestation and
cultivation) (Bronstert et al., 2002). Compaction caused by, for example, heavy agricultural machinery
describes the densification of soil particles and the loss of pore spaces. It will lead, for instance, to a decrease
in Ks and a reduced infiltration capacity. Surfaces sealed by physical structures such as roads or buildings
will make the ground impermeable for water and favour runoff generation. Hence, runoff occurs either
when (I) the surface is impermeable, (II) the precipitation intensity exceeds the infiltration rate, or (III) the
soil is fully saturated.
A conventional way to foster understanding of complex surface and subsurface processes is hydrological
modelling (Raudkivi, 1979). In the literature, three main types of hydrological models are identified, (I)
empirical models, (II) conceptual models, and (III) physical based models. Empirical models, on the one
hand, are observation-based, data-driven models, conceptual models, on the other hand, consider all
components of the hydrological process and work with semi-empirical equations. Whereas physical models
are based on mathematical equations to represent real-world processes as realistic and simple as possible
(Devia & Ganasri, 2015; Merritt et al., 2003). They can provide a variety of information and can be applied
to a wide range of applications such as flood process modelling, vegetation growth modelling, and
groundwater modelling.
According to Bout and Jetten (2017), physically based models can be further classified as decoupled models
and integrated watershed models. The former separates upstream runoff generation from flooding in the
downstream areas. While an upstream model is used to generate discharge values, a downstream model is
then responsible for the flood simulation. The latter simulates the hydrology at the watershed scale and
generates runoff based on rainfall and infiltration. Bout and Jetten (2017) further pointed out that decoupled
models are often not applicable for flash flood modelling as flash floods are not necessarily linked to an
overflow of a channel since they are also generated in adjacent sloping terrain.
Widely used integrated catchment models are for instance MIKE-11 (DHI, 2017), the Hydrologic Modelling
System HEC-HMS (Scharffenberg, 2016), the Soil and Water Assessment Tool (SWAT) (Shekhar & Xiong,
2008) and the Limburg Soil Erosion Model (openLISEM) (Bout & Jetten, 2018). A distinction can be made
considering input data requirements, the processes modelled, or the temporal scale they operate in. Where
MIKE-11, SWAT, and HEC-HMS, for example, can work on a temporal scale of years, they incorporate
evapotranspiration and groundwater flow. OpenLISEM, on the other hand, is a purely event-based model
working with single rainfall events and therefore neglects such processes. Operating in time steps of minutes,
openLISEM is tailored to model flash flood processes.
The drawback of physically based models is their immense data demand and the output dependency on the
input data quality (Sanchez-Moreno et al., 2014). OpenLISEM requires a minimum of 24 maps, which can
be derived from four main sources encompassing rainfall, topography, surface and soil related information
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
3
(Bout et al., 2018). Among others, one characteristic of physically based hydrological models is, that soil
data is an essential baseline data, and the model output is highly dependent on its quality (Merritt et al.,
2003).
Usable soil data for any kind of application are rare and substantial investments have to be taken in new
detailed soil measurements (Sanchez et al., 2009). Therefore, it can be assumed that especially developing
countries, which are usually most prone to natural hazards and the effects of climate change (IFRC, 2016),
exhibit a deficit of appropriate soil data. Existing soil data are often inaccurate, outdated, with coarse spatial
resolution, disregard soil properties, as well as site-specific geomorphological processes due to inadequate
methods (Arrouays et al., 2014; Grunwald et al., 2011; McBratney et al., 2003). In the light of that, the data
are not valid to support efforts of sustainable development, which is gaining importance in the countenance
of growing pressure on the planet, caused by climate change, biodiversity loss, land degradation and
urbanization (FAO & ITPS, 2015; Montanarella & Vargas, 2012).
Recently, the International Soil Reference and Information Centre (ISRIC) made ‘SoilGrids’ (SG), a global
soil dataset publicly available to bridge the gap between soil data demand and availability (Hengl et al., 2015).
SG is based on machine learning and comes in a 250 m grid resolution. According to Hengl et al. (2017),
SG spatial predictions are based on approximately 150,000 individual soil profiles spread over the world,
which had to be merged. Standardisation methods were applied to translate soil data provided in national
classification systems (up to 20 %) to the international classification systems (World Reference Base and
USDA). For areas with no existing observation points (e.g., mountain tops, steep slopes and inaccessible
tropical rainforest), expert-based pseudo-observation were input. Remote sensing soil-covariates such as
MODIS land products (e.g., land cover, surface temperature and Enhanced Vegetation Index) and Shuttle
Radar Topography Mission (SRTM) digital elevation model (DEM) derivatives (e.g., slope, profile curvature,
and valley depth) provide additional support for the predictions (Hengl et al., 2017). Since it is estimated
that detailed soil profiles are available for only one-third of the world (Bonfante & Bouma, 2015), the
significance of global digital datasets like SG is increasing to overcome existing data gaps.
So far, there has been no initiative in the published literature to verify SG performance and accuracy in
disaster risk research. Nevertheless, SG was deployed in several studies. Bout and Jetten (2017) used it to
validate flow approximations in hydrological modelling, Shrestha (2014) for flash flood modelling in the
Fella basin in Italy and Chen et al. (2016) in a multi-hazard risk assessment. Further applications can be
found in disciplines like crop modelling (Han et al., 2015) and studies concerning carbon stock estimations
(Tifafi et al., 2018). However, none of these studies assessed how SG may have influenced the study outputs
with its prediction-based data points, and coarse spatial resolution and thereby potentially not reflected
spatial variability. None of them discussed the limitations of SG nor conducted an accuracy assessment.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
4
1.2. Problem statement
Soil data is essential baseline information for hazard modelling and therefore, also for flash flood modelling.
In many parts of the world, affected by flash flood hazards, detailed hazard assessments are needed to
support sustainable development. Though, detailed soil data are still lacking, especially in remote areas. The
global digital soil dataset SG, developed by ISRIC, represents a possibility to bridge the gap between soil
data demand and availability, but its performance has to be tested. Besides this, it is evident that flash flood
frequencies and intensities are increasing, due to climatic alterations and due to socio-economic
developments and the accompanied land use and land cover changes. Therefore, in this thesis, an impact
assessment of soil information on flash flood modelling was performed, to compare the performance of
different soil information sources and to identify potential limitations of SG data. Furthermore, the effects
of long-term land cover changes were investigated. For this purpose, the integrated flood model
openLISEM was deployed, (I) with SG data, (II) with detailed field data (FD), and (III) with changed land
cover information. The respective model outputs provide information regarding SG and the effects of land
cover changes. A better understanding of SG limitations is expected to increase both its attractivity for the
research community, governmental bodies and national and international organizations working in the area
of disaster risk reduction, and cautiousness when used. In addition, with the knowledge about the influence
of land cover changes on flash floods, risk-informed decision making, and sustainable spatial planning can
be promoted. In conclusion, this thesis is aligned with the following global goals and agendas, to make a
meaningful contribution:
Transforming our World - the 2030 Agenda for Sustainable Development - Particularly the
following Sustainable Development Goals (SDGs):
• SDG 1: End poverty in all its forms everywhere; Target 1.5: By 2030, build the resilience of the poor
and those in vulnerable situations and reduce their exposure and vulnerability to climate-related
extreme events and other economic, social and environmental shocks and disasters.
• SDG 13: Take urgent action to combat climate change and its impacts; Target 13.1: Strengthen
resilience and adaptive capacity to climate-related hazards and natural disasters in all countries.
The Sendai Framework for Disaster Risk Reduction (SFDRR):
• Priority 1: Understanding disaster risk
• Priority 3: Investing in disaster risk reduction for resilience
• Priority 4: “Build Back Better” in recovery, rehabilitation, and reconstruction
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
5
1.3. Objectives and research questions
The main objective of this research is to conduct an impact assessment of soil information and land cover
change on flash flood modelling on a watershed scale in Thailand. To achieve the main objective, the
following specific objectives with corresponding research questions were set:
Objective 1: Comparative analysis of SoilGrids (SG) versus field data (FD) for flash flood modelling;
1. How well do SG and FD correlate?
2. How do the soil properties of (I) FD, and (II) SG relate to the main land cover types?
3. How do the soil properties of (I) FD, and (II) SG relate to the terrain?
4. Is model calibration based on historical flood marks from a nearby watershed possible?
5. What are the quantitative differences of the model output using, (I) FD, and (II) SG in
relation to flood dynamics?
6. What is the sensitivity of the flood dynamics to different soil depths using, (I) FD, and (II)
SG?
Objective 2: Analysis of the effects of land cover change on flash flood behaviour;
1. Which land cover changes occurred in the study area between 2005 and 2018?
2. What are the possible reasons behind these land cover changes?
3. Which land covers generate the highest average runoff?
4. How do these land cover changes affect runoff generation and flood dynamics?
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
6
2. RESEARCH AREAS
The research was carried out in two areas that are located in Uttaradit province in Northern Thailand within
the Latitudes 17o37’N – 17o52’N, and Longitudes 99o55’E – 100o09’E (Figure 1). The first is Ban Da Na
Kham watershed (86.9 km2), and the second Laplae watershed (156.9 km2). Two watersheds were selected
as in Ban Da Na Kham watershed diverse types of land cover changes were observable, whereas Laplae
offers the possibility of model calibration as flood marks from a flash flood event in 2006 are available. Both
sides are located in the vicinity to each other where the former is located in Mueang district and the latter
in Laplae district (Figure 1). In general terms, Ban Da Na Kham watershed is tube-shaped, whereas Laplae
is more elongated. However, even having different shapes, Laplae is expected to be suitable for calibration
purposes since, geology, land cover and topography are nearly identical in both watersheds.
Topography and Land Cover
Most areas in the watersheds are characterised by mountainous terrain crisscrossed by small valleys. Highest
areas are in the northern parts of both watersheds with elevations up to 828 m above mean sea level in
Laplae and up to 754 m in Ban Da Na Kham. The lowest areas are in the southern parts of the watersheds
with a minimum of 37 m in Laplae and 66 m in Ban Da Na Kham.
A variety of land cover types ranging from urban areas over orchards to natural forest are present in both
watersheds. Sloping terrain and narrow valleys are cultivated with fruit trees e.g. long kong, banana and
durian trees, whereby larger valleys are used for paddy rice cultivation. Mountain summits and steep slopes
Thailand
Figure 1. Location map of Ban Da Na Kham and Laplae watershed.
Ban Da Na Kham
Laplae Da Na
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
7
are covered by impassably natural forest. Furthermore, teak plantations for commercial timber production
can be found in different places within Ban Da Na Kham watershed. In the southern lowland area of Laplae
watershed, the city Si Phanommat with a population of 3200 is situated. On the contrary, Ban Da Na Kham
is sparsely populated, having small isolated villages lined along the main roads of the watershed.
Geology
From a geological point of view, both the watersheds have three distinct geological formations. In the higher
altitudes in the northern parts, they consist of the Khao Ploung formation comprising shale or mudstone
interbedded with greywacke sandstone. In the south, the Lab Lae formation takes over occupying the
majority of both watersheds with sandstone (greywacke) interbedded with shale. Alluvial deposits containing
gravel, sand, silt and clay can be found in the southern lowland areas (Annex 1).
Climate
Uttaradit province has a humid tropical climate influenced by the north-eastern and south-western Monsoon
which determines the three seasons, namely cold (November – February), hot (February – May) and rainy
(May – October). Annual precipitation ranges from approximately 830 to 2.100 mm, with August being
usually the wettest month of the year. The warmest month is April with an average temperature of 30 °C
and the coldest average temperatures with 24 °C is in January (Figure 2).
Historic flash flood event
On 22nd of May 2006 flash floods, landslides, and debris flows were triggered by prolonged heavy rainfall in
several provinces in lower Northern Thailand. Among others, Uttaradit province was affected by this
unusual event. Flood heights of several meters were reported among different communities in the province.
In the wake of the disaster, 87 people died, 700 houses were totally damaged, and almost 4000 partially.
Additionally, more than 370 hectares of mountainous mixed-fruit tree orchards were completely destroyed
due to landslides, and some hundred hectares of lowland orchard and cropland were flooded and buried by
mud (Boonyanuphap, 2013). These consequences are due to three days of consecutive rainfall with a total
of approximately 400 mm, having its peak on May 22, 2006, with a total rainfall amount of 263.7 mm
recorded at the meteorological station in Uttaradit city.
10
15
20
25
30
35
0
50
100
150
200
250
300
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Tem
per
atu
re (
°C)
Pre
cip
itat
ion
(m
m)
Avg. Precipitation (mm) Avg. Temperature (°C)
Figure 2. Climate diagram of the research area. Based on data from the Thai Meteorological Department for Uttaradit city station (ID: 351002) for the period 2006 to 2018.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
8
In general, the northern part of Thailand is frequently affected by flash floods and landslides (CFE-DM,
2018). However, the event in 2006 marks a historical event in Thailand’s history because it took the lives of
many people and caused tremendous economic damage. Thus, this event was selected to serve as the basis
for the modelling efforts in this research.
3. METHODOLOGY
This chapter gives first an overview of the data used in this research project, and second describes the
methodologies applied. The methodology used to satisfy the research objectives is divided into three parts.
The first part consists of the methodological steps undertaken to do the soil data analysis and to further
conduct a comparative assessment between FD and SG. The second part has a strong remote sensing
component, focusing on Google Earth Engine (GEE) and machine learning algorithms in order to conduct
land cover mapping and land cover change analysis. In the third part, the soil data and land cover
information are combined and used as an input for the hydrological modelling.
3.1. Data
The data used for this research study is summarized with its sources and properties in Table 1.
Table 1. Data used and its source and properties.
Type Method Spatial Resolution Source
Alos Palsar Digital DEM Satellite (Radar) 12.5 m Alaska Satellite Facility, 2006-2011 (www.asf.alaska.edu)
SRTM DEM Shuttle Radar 30 m U.S. Geological Survey (www.usgs.gov)
Digital Soil Data Random Forest classification 250 m Hengl et al. (2017) (soilgrids.org)
Soil Data Undisturbed and disturbed sampling Point Fieldwork
Rainfall Rain gauge (daily measurements) Point Thai Meteorological Department (1952-2018)
Land Cover Maps Landsat 5 and 8 Random Forest classification
30 m -
Road Network - Vector OpenStreetMap (www.openstreetmap.org)
Historical flood height measurements
Field measurements Point Fieldwork
3.2. Soil data analysis
In this sub-section, the methodology to answer the following research questions is outlined:
• How well do SG and FD correlate?
• How do the soil properties of (I) FD, and (II) SG relate to the main land cover types?
• How do the soil properties of (I) FD, and (II) SG relate to the terrain?
The first part comprises statistical methods for the assessment of (I) the FD, and (II) the global digital soil
dataset SG (250 m) in relation to land cover types and the terrain. The second part is dedicated to the
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
9
methodology of the comparative analysis between the two datasets. In the third part, the soil sampling
strategy, the sampling process and the methods used for the laboratory analysis of the soil data are described.
Lastly, the generation of a slope unit map using the tool geomorphon (Stepinski & Jasiewicz, 2011) is
explained.
3.2.1. Pedotransfer functions
Pedotransfer Functions (PTFs) are regression equations established mostly based on large soil datasets and
are used to predict SHPs (Looy et al., 2017). For predictions, soil texture (sand, silt and clay contents) and
bulk density (Db) are commonly required (Williams et al., 1992). Some PTFs make also use of soil organic
matter (SOM) for example, see Wösten et al. (1999), Nemes et al. (2005) and Saxton & Rawls (2006). PTFs
generally reflect the interactions of soil properties from the soils they were constructed of, and hence, their
prediction capabilities are coupled to the underlying soil database (Nemes et al., 2005). PTFs for the humid
tropics and especially for Asia are limited in number according to a review article by Botula et al. (2014).
Often, they are built based on small or even unknown sample sizes, which potentially introduce a large
uncertainty. For this study, the PTFs of Saxton & Rawls (2006) were used to estimate SHPs of SG data. The
PTFs are based on a combination of nonlinear multiple regression equations with texture and SOM,
combined with the hydraulic conductivity and water retention equations of Brooks and Corey (1964). Those
PTFs were originally created from approximately three thousand USA soil samples and represent, according
to Gijsman et al. (2003), the most accurate PTFs compared to other common methods. The conversion
from SG physical and chemical properties to SHPs was done using the PCRaster script from Jetten and
Shrestha (2018) (Annex 2).
3.2.2. Independent soil assessment
For the two soil datasets used in this study (FD and SG), an independent assessment was conducted. This
assessment included the calculation of descriptive statistics such as the mean and standard deviation for
statistical data summarization and measure of data dispersion. Furthermore, Box and Whisker charts were
plotted for visual investigation of the data variability within the landscape.
To foster an understanding of the soil-landscape relationship, the soil data were grouped based on the land
cover class and based on their slope position (explained below). Moreover, selected pairwise correlations of
soil properties were conducted using Pearson’s coefficient r, which is a measure of linear interdependency
of two variables (Eq. 1). This was done to investigate statistically dependencies. The coefficient ranges from
-1 to +1, with -1 being an indication of a negative dependency, 0 describes independency, and +1 a positive
dependency. The computations were done with the software package IBM SPSS statistics 25. Equation 1
describes Pearson’s r.
𝑟 =∑(X − X̅)(Y − Y̅)
√∑(X − X̅)2 √∑(Y − Y̅)2 (1)
Where X and 𝑌 are the two variables of interest and �̅� and �̅� are the respective mean values of the
variables.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
10
3.2.3. Comparative soil assessment
Comparative soil assessment was conducted by using Cosine Similarity and Wilcoxon Signed-Ranked test.
The Cosine Similarity is usually applied as a similarity measure for text documents by creating frequency
vectors of keywords (Han et al., 2013). In the frame of this research, the frequency vectors are substituted
by soil property vectors. Computed is the cosine of the angle between two vectors. According to Han et al.
(2013) a cosine of 0 does mean that the two vectors have a 90 degrees angle, and therefore are orthogonal
to each other, meaning no match between the two vectors. The closer the cosine value is to 1, the more
similar the vectors are. The advantage of cosine similarity is that the vectors can accommodate as many
variables as needed. Therefore, different combinations of soil properties can be tested. Interpretation of the
results is comparable to Pearson’s r. Equation 2 defines the cosine similarity.
cos(𝑋, Y) =X • Y
‖𝑋‖‖𝑌‖=
∑ 𝑋𝑖𝑌𝑖𝑛𝑖=1
√∑ 𝑋𝑖2𝑛
𝑖=1 √∑ 𝑌𝑖2𝑛
𝑖=1
(2)
Where X and Y are two vectors of comparison, ‖𝑋‖ and ‖𝑌‖ are the Euclidean norms, or the length of the
vectors. As soon as the soil property vectors are composed of soil properties with different units, e.g. Db (g
cm-3) and Ks (mm h-1), a comparison becomes difficult. Therefore, the normalisation of the measurements
is a necessary data preparation step. Data normalisation was done with min-max normalisation using
equation 3 (Han et al., 2013).
𝑋 − 𝑋𝑚𝑖𝑛
𝑋𝑚𝑎𝑥 − 𝑋𝑚𝑖𝑛 (3)
Where 𝑋 is the soil property of interest; 𝑋𝑚𝑖𝑛 and 𝑋𝑚𝑎𝑥 represent the minimum and maximum of the
measured soil property, respectively. Normalisation results are given in a sequence of numbers ranging from
1 representing the maximum measured value to 0 representing the minimum measured value.
As a second similarity measure, the Wilcoxon Signed-Rank test was conducted. This statistical measure was
chosen above the common Paired-Samples t-test because (I) not all of the tested variables (soil properties)
followed the assumption of normal distribution, and (II) because the Wilcoxon Signed-Rank test is robust
to the effects of possible outliers. Calculated differences were considered to be significant if the z-value was
less than the critical P-Value of 0.05 (Field, 2009).
The normality of the data was investigated with the Shapiro-Wilk test, kurtosis and skewness. If the skewness
and kurtosis z-value (P>0.05) is within the span of -1.96 to +1.96 and a Shapiro-Wilk P-Value above 0.05,
a normal distribution can be expected. If one of the conditions is violated the data can be expected to be
non-normal distributed (Field, 2009).
3.2.4. Soil sampling strategy
According to Möller et al. (2008), landscapes are determined by their landforms such as flood plains, alluvial
fans, ridges and slopes. The assumption that similar landforms in an area exhibit similar soils is widely spread
in the soil surveying community (McKenzie et al., 2008; Möller et al., 2008; Schaetzl & Anderson, 2005).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
11
The rationale behind this is the assumption that the origin of landforms is controlled among others by
geology and the prevailing surface and subsurface processes and that the underlying geology forms mostly
the parent material of the occurred soils. Hence, landforms are expected to give an inference on soil genesis,
soil formation and the forming processes (Möller et al., 2008). If complying with this line of thinking, it
would be obvious to either base the soil sampling strategy on an existing geological or a soil map, or to
conduct a terrain analysis dividing the area into its terrain units. As no such maps were available prior to the
fieldwork and explicit terrain units could not be identified, another approach had to be chosen.
Therefore, the hillslope concept which divides the landscape into five slope positions (Schoeneberger et al.,
2012) (Figure 3), was considered to represent an acceptable alternative to the other approaches for a soil
sampling strategy. Slope positions are known to serve as a good predictor of soil properties and represent a
comparable simple way of dividing the landscape into different units (Miller & Schaetzl, 2015; Schaetzl et
al., 2005).
To aid the selection of suitable soil sampling sites, the elevation profile tool of Google Earth Pro was used.
By creating cross sections, an identification of the approximate slope positions and placing of potential
sampling points was possible. In the process, attention to accessibility, land cover type and spatial
distribution was given. The proximity to roads and paths was expected to ensure accessibility to the soil
sampling sites. Since the sampling sites were chosen based on the hillslope position and the land cover types,
the sampling scheme was purposive.
3.2.5. Soil sampling
Maintaining the pre-selected sampling points proved difficult after the first investigation of the watershed.
Accessibility issues due to steep slopes, private property or difficult paths required an on-site selection of
sampling locations (Annex 2.1). However, undisturbed and disturbed soil samples were taken on 48
locations within the land cover mixed forest, teak plantation, orchard (banana and long kong) and cropland
(corn and bean). At each sampling site, GPS coordinates were recorded with a Garmin GPS receiver. The
undisturbed core samples were collected at a depth of 5 cm with a steel core sampler having a diameter of
5 cm. In addition, soil depth was measured at each location with an auger (up to 1 m). Since the terrain of
the watershed is characterised by steep slopes, narrow valleys and narrow hilltops, the slope positions to
sample were restricted to summit, backslope and valley. A summary of the sampling points can be found in
(Table 2).
Figure 3. Slope positions in a natural landscape; adapted from Schoeneberger et al. (2012).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
12
Table 2. Soil samples per slope position and land cover.
Slope Position
Lan
d C
over
Summit Backslope Valley
Orchard 3 19 12
Cropland 1 - 3
Teak Plantation - 1 1
Mixed Forest 2 4 2
3.2.6. Laboratory analysis
The collected soil samples were analysed in the laboratory facilities of the Faculty of Engineering of the
Naresuan University in Phitsanulok, Thailand. Ks, Db, and porosity were determined based on the
undisturbed surface core samples. For measuring PSD and SOM content, the disturbed samples were used.
Details on the results of the laboratory analyses for each measured soil physical and chemical property along
with their location information are given in Annex 2.2 and Annex 2.3.
Ks was measured using the Constant Head method as described in the operating instructions for a
laboratory-permeameter (Eijkelkamp, 2013). Subsequent to the Ks measurement, Db and porosity were
measured following the method introduced by Soil Survey Staff (2014). Using the Hydrometer Method as
published by the Soil Science Society of America and outlined in the Soil Survey and Laboratory Manual
(Soil Survey Staff, 2014), the PSD was assessed. For the determination of the SOM content of the disturbed
surface samples, the Loss-on-Ignition (LOI) method as set out by Schulte and Hopkins (1996) was applied.
LOI was chosen as it represents an even more effective and simpler way to determine SOM compared to
other conventional methods such as the Walkley and Black Carbon (WB-C) method, which requires
additional chemicals and laboratory facilities (Paramananthan et al., 2018). Detailed descriptions of the
methods and equations used can be found in Annex 2.4.
3.2.7. Slope unit delineation
For the analysis of the soil-landscape relationship, a slope unit map was generated using the GRASSGIS
extension Geomorphon. According to Jasiewicz and Stepinski (2013), Geomorphon is a DEM based pattern
recognition approach meant to be used for classification and mapping of landforms. Characterisations are
done by the use of a local ternary operator that assigns an 8-tuple pattern making use of the symbols “-“,
“0” and “+”, which describe a neighbouring cell as lower, equal or higher than the focus pixel. As the
method is based on the line-of-sight principle, the neighbours are not necessarily the immediate neighbours,
rather depended on the search radius L (user-defined) which varies based on the scale of interest. Where a
Figure 4. Common landforms recognisable with Geomorphon (Jasiewicz & Stepinski, 2013).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
13
smaller value for L detects landforms locally and, a larger search radius L will detect landforms on several
sizes. Further parameters which can be defined by the user are the flatness threshold and skip radius
(Jasiewicz & Stepinski, 2013). Using a greater value for the flatness threshold and skip radius will result in
an extent of the delineated plains, by reducing the influence of small irregularities (Jasiewicz et al., 2013;
Veselsky et al., 2015). For simplification, Geomorphon is limited to recognize the 10 most commonly
landforms, namely: flat, peak, ridge, shoulder, spur, slope, hollow footslope, valley and pit (Figure 4).
In this study, different input values for L and the skip radius were tested and compared to the slope positions
observed in the field, in order to determine the best combinations suitable for the research area and DEM
resolution used. As the terrain specifications restricted the observable landforms to summit, backslope and
valley (as discussed above), the output of Geomorphon was aggregated to those three units. Hence, summit
and ridge were combined to form the summit unit. Shoulder, spur, slope and hollow as well as footslope
are categorized as backslope unit, and lastly, the valley unit represents a combination of valley, depression
and flat Geomorphons.
3.3. Land cover mapping
In this section, the methodology is described which shall satisfy the following research questions:
• Which land cover changes occurred in the study area between 2005 and 2018?
• What are the possible reasons behind these land cover changes?
Land cover maps for the years 2005 and 2018 were produced using the computational planetary-scale
geospatial analysis platform GEE. The year 2005 was chosen as reference year as it represents the research
area with its land cover before the great flash flood event in May 2006. As the comparative year, 2018 was
selected since it represents the state of the area as observed during the fieldwork. GEE as a tool was used
for the classification as it simplifies and accelerates access and analysis of remote sensing data for land cover
classifications. GEE is a cloud-based online-platform accessible from any place in the world; the only
requirements are a functional computer and a stable internet connection. Using the computational resources
of Google’s Data Center Infrastructure and having petabytes of remote sensing datasets such as Landsat,
Sentinel, Modis and non-satellite imagery, it enables parallel and fast computations on large datasets. Details
on GEE can be found in Gorelick et al. (2017). Presented procedures in this chapter are based on the
Figure 5. Workflow of the land cover classification; adapted from Saah et al. (2019).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
14
Regional Land Cover Monitoring System (RLCMS) developed by SERVIR-Mekong as extensively described
by Saah et al. (2019). Several adaptations and simplifications were done to tailor the methodology to the
scope of this research. Figure 5 shows the overall workflow of the land cover classification.
3.3.1. Random forest classifier
Random forest (RF) was used as supervised classification algorithm because it allows higher mapping
accuracies in comparison to conventional classifiers (e.g., maximum likelihood or simple decision tree)
(Rodriguez-Galiano et al., 2012). Conventional image classification techniques often only exploit the spectral
signatures and sometimes texture or pattern for pixel discrimination (Domaç & Süzen, 2006). Using RF, an
enhanced differentiation between different land cover classes even in complex terrains is possible due to the
incorporation of ancillary variables (Tsai et al., 2018; Rodriguez-Galiano et al., 2012). Variables can be for
example spectral band indices (e.g., Normalised Difference Vegetation Index (NDVI) or Normalised
Difference Water Index (NDWI)) and terrain derivates like slope, aspect and elevation. When adding such
variables, additional contextual information can be included in the classification process to improve the
discrimination between different land cover classes (Domaç et al., 2006). RF found its successful application
in various land cover studies for example, in Colditz (2015), Nguyen et al. (2018) and Steinhausen et al.
(2018).
According to Breiman (2001), the algorithm creates a ‘forest’ by growing a number of n (user-defined)
decision trees based on a random selection of a subset of data samples and a random selection of a subset
of n variables. Being an ensemble classifier, RF creates the random subsets of the training data by using
bootstrap aggregation (bagging) procedure (Figure 6). The output sample subsets are called bootstrap
samples. Bootstrap samples are created on a random basis with the option of replacement (bagging). Thus
each selected sample can be used more than once, once, or not at all within one subset (Breiman, 2001).
Figure 6. Random Forest classification; V = random variables and S = random samples.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
15
Subsequently, different classifiers (trees) are built on the subsets of data, which are all a bit different even
though they were created from the same original dataset. Then the final classifier is built by averaging the
predictions of all sub-classifiers, in doing so it reduces the variance by taking out complexity. Hence, the
stability and accuracy of the algorithm improves in the prediction making process (Bauer & Kohavi, 1999).
In other words, in RF is a pixel classified based on a set of n variables. According to Breiman (2001), the
forest trees are independently grown by nodes which represent tests of these variables. The algorithm
computes the information gain contributed by all the variables and splits the node accordingly with the best
split contributed by the variable with the highest information gain. Each branch of a tree represents
accordingly the decision of a test. Finally, each grown tree votes for a class and the class with the most votes
(majority vote principle) is chosen for the respective pixel (Breiman, 2001) (Figure 6). Overall, the
effectiveness of a random forest compared to ordinary decision trees lies in its variety due to bootstrapped
sampling and the use of random subsets of variables at each step.
3.3.2. Land cover typology
Defining a land cover typology is a crucial step for any land cover mapping activities and is driven by its
purpose and the user needs. As such, a typology is created to describe the observable (bio) physical cover
of an area in a comprehensive and concise manner. By following objective criteria for the definition of clear
and precise classes, various land covers can be discriminated without redundancy and overlap. This supports
the observation of land characteristics and the monitoring of their dynamics. Furthermore, it enables the
comparison of different land cover products and the analysis of land cover changes (Faber-Langendoen et
al., 2009; Vegetation Subcommittee, 1997).
The land cover typology applied in this research study is adapted from the RLCMS. The RLCMS, in turn,
is based on the Land Cover Classification System (LCCS) and associated tool developed by the Food and
Agriculture Organization of the United Nations (FAO) (Di Gregorio, 2005, 2016; Di Gregorio & Jansen,
2000). The LCCS was launched with the fundamental idea to contribute to the establishment of an
international land cover classification standard. Its approach is physiognomic-structural, meaning the
elements are described based on their overall appearance and spatial distribution pattern (Di Gregorio,
2016). The definitions of the different land cover classes are specified below:
• Cropland represents a combination of irrigated or flooded rice paddy fields and dryland crops such
as corn and vegetables. The majority of the cropland is covered by rice paddies which are intensively
tilled (more than one cycle per year), and even being flooded, rice makes up the majority of
observable surface cover. Ideally, irrigated crops and dryland crops should be classified separately
as they exhibit different characteristics. However, in this research study, a separation was not
possible as the area covered by dryland crops was insufficient to represent an individual class.
Woody perennial crops such as fruit trees are excluded from this class (Figure 7).
• Orchards are defined by long term occupation of land by perennial crop trees (Blanchez, 1997). In
the study area, they are mainly represented by long kong and banana plants. The fruit trees are
planted in a line pattern with gaps of several meters in between. Ground cover is dense but shallow
grass (Figure 7).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
16
• Mixed forests include evergreen broadleaf and deciduous trees (D’Souza, 2000). The understory
vegetation is very dense (Figure 7).
• Teak plantations consist of deciduous teak trees (Tectona grandis). Once a year the trees carry white
flowers, arranged in dense clusters. Furthermore, the plantations are arranged in a line pattern,
recognisable especially in young plantations. In mature plantations the understory vegetation is
pronounced (Figure 7).
• Urban and built-up areas were defined as areas covered by human-made structures (e.g., buildings
and roads) (D’Souza, 2000) (Figure 7).
• Water bodies were defined as surface water open to the sky and containing fresh water (Pekel et
al., 2016) (Figure 7).
3.3.3. Image processing
Surface reflectance products of U.S. Geological Service (USGS) Landsat 5 Enhanced Thematic Mapper
(ETM) and Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) were used to
create composite images for the years 2005 and 2018. ETM images include visible bands, two shortwave
infrared bands as well as one thermal and one near infrared band. OLI and TIRS products provide visible
bands, two short wave infrared bands, as well as two thermal bands and one near infrared band. Products
from Landsat 7 ETM+ were excluded due to the Scan Line Corrector failure, which occurred after 2003.
Images were acquired for two timeframes. For 2005 the timeframe was defined from 1st of January 2005
until 20th of May 2006. This period was chosen as the flash flood event to be modelled happened on the
22nd of May 2006, and modelling efforts require a land cover map representing the condition of the area
db)
da dc)
dd de df
Figure 7. Land cover classes in the research areas. a) Cropland, b) Orchard, c) Teak Plantation, d) Mixed Forest, e) Urban, f) Water Bodies (Google Earth Pro, 2019).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
17
before the event occurred. However, just two Landsat 5 images with insufficient quality (blurred) were
available prior to this date in the same year. Hence, images from 2005 were included in the timeframe.
For 2018, enough images were available to create seasonal composites. The advantage of having seasonal
composites is that seasonal variability of the vegetation is captured which adds additional information to the
classification procedure later on (Flood, 2013; Rodriguez-Galiano et al., 2012). Therefore, seasonal
composites for winter, summer and rainy season were generated. In the GEE code of Saah et al., (2019) the
seasons for Thailand are defined as follows using calendar days:
• day 305 - 59 (+1 year) is defined as ‘dry cool’ season or winter;
• day 60 - 179 is defined as ‘dry hot’ season or summer; and
• day 180 - 304 (+1 year) is defined as ‘rainy’ season
Details all on images used for the composite of 2005 and the seasonal composites of 2018 can be found in
Annex 3.
Atmospherically corrected orthorectified surface reflectance data products by USGS Landsat 5 and 8 are
available in the GEE data archive. Images from Landsat Mission 5 have been atmospherically corrected
using the LEDAPS algorithm (Schmidt et al., 2013) including a cloud, shadow, water and snow mask
produced by CFMASK (Zhu & Woodcock, 2011). Landsat 8 images have been atmospherically corrected
through the application of the Landsat Surface Reflectance Code as described by Vermote et al. (2016).
Following the image pre-processing steps of the SERVIR-RLCMS method (Saah et al., 2019), the data were
subjected to further corrections to take into account for distortions caused by sensor, solar, atmospheric,
and topographic effects (Young et al., 2017). Therefore, cloud and cloud shadow removal algorithms, as
well as the bidirectional reflectance distribution function (BRDF), and topographic corrections have been
applied. In Annex 3.1-3.4 GEE codes for image processing adapted from Saah et al. (2019) can be found.
Cloud and cloud shadow masking
The clouds were masked using the pixel-qa band and Google’s cloudScore algorithm (Chastain et al., 2019).
In general, cloudScore takes advantage of the spectral and thermal properties of clouds. It locates and
removes pixels which are cold, and bright in all visible and infrared bands. To differentiate between clouds
and snow, the Normalised Difference Snow Index (NDSI) was calculated (Salomonson & Appel, 2004).
Chastain et al. (2019) provides details on the algorithm. In this study, a cloud score threshold of 20 was
applied as it delivered the best results.
Cloud shadows were removed by utilising the Temporal Dark Outlier Mask (TDOM) (Housman et al.,
2015). The algorithm detects dark pixels in the infrared bands (NIR, SWIR1, and SWIR2) by doing statistical
outlier analysis. Pixels are identified as cloud shadow if they have not always been dark in previous or
following images. If no uncontaminated pixel from previous or later images could be found as a replacement
for a masked cloud or cloud shadow pixel, gaps with no data appeared. In that case, gap filling was done
with pixels extracted from a composite of the previous year. The codes for cloud removal are shown in
Annex 3.1 and for shadow removal in Annex 3.2.
BRDF correction
Surface reflectance measurements of Landsat images are subjected to directional effects. Those are caused
by variations in solar zenith angle over time, and additional through different angles of vision during the
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
18
time of acquisition (Roy et al., 2016). The BRDF attempts to describe the directional reflectance (Roy et al.,
2017). The method of Roy et al. (2016) was used for correction. GEE codes are available in Annex 3.3.
Topographic correction
Analysing any kind of remotely sensed imagery in respect to land cover types is challenging and not effective,
without topographic corrections. Same objects on Earth happen to exhibit different reflectance values
depending on their location in the terrain. Especially in mountainous areas terrain slope and aspect cause
variations in illumination angle and reflection geometry (Colby, 1991). Compensation of those different
solar illuminations can be achieved by topographic corrections (Riaño et al., 2003). In the literature, various
correction approaches exist. The cosine correction by Smith et al. (1980), for example assumes that the total
irradiance at a pixel is directly proportional to the cosine of the incidence angle. Hence, this model is applied
when illumination is not originating from the zenith. On the other hand, the Sun-Canopy-Sensor Correction
(SCS) model enhances the cosine correction by normalising the radiances from the sunlit canopy (Gu &
Gillespie, 1998). However, the Modified SCS Topographic Correction method, also known as SCS+C
(Soenen et al., 2005) was used which accounts for diffuse radiation caused by atmospheric and terrain
sources and adjusts for terrain and at the same time preserves the sun-canopy geometry. Codes for
topographic correction are available in Annex 3.4.
Composite assembling
Various methods exist for image compositing based on multiple images captured on different dates (Flood,
2013). For this study, median compositing was used. In median compositing, the median over a time series
of images represented by an image collection of the defined timeframe is calculated. The output is an
individual image with each pixel having the median value out of all images. By creating such a composite, it
is possible to generate a representative image for the timeframe of interest without or with minimal noise
caused by for example, clouds or cloud shadows (Flood, 2013).
3.3.4. Covariate layers
As already mentioned above, adding variables (covariates) to the classification enables the incorporation of
contextual information to the classification procedure, which can improve the prediction of land cover
classes (Cingolani et al., 2004; Domaç et al., 2006; Maselli et al., 1995; Rodriguez-Galiano et al., 2012).
Covariate layers were mainly generated by exploiting the already available spectral bands of the composite
images as it represents a simple, effective, and reliable form of adding contextual information (Boonprong
et al., 2018). Therefore, Landsat median composite derivatives were calculated as covariates. Additionally,
terrain derivatives extracted from the freely available SRTM digital elevation dataset (Farr et al., 2007) were
added. In the course of this section, all included covariate layers are briefly explained, and reasoning for
their incorporation is provided. The adapted GEE codes from Saah et al. (2019) for the covariate
implementation are available in Annex 3.5 (seasonal composite) and Annex 3.6 (yearly composite).
Landsat derivatives
The bands of the median composites were used to calculate numerous normalised differences (NDs) and
other more complex indices. The ND vector (Eq. 4) (Angiuli & Trianni, 2014) finds its application in image
classification in order to extract information from spectral bands by removing errors caused amongst others
by differences in space, time and acquisition (Angiuli et al., 2014).
ND(Band1,Band2)=Band1-Band2
Band1+Band2 (4)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
19
Calculations comprised well known normalised indices such as the NDVI (Jeong et al., 2017), the NDWI
(McFeeters, 1996) and the Normalised Burn Ratio (NBR)(Hislop et al., 2018). In Table 3, all applied band
combinations for NDs are shown. Each of the indices was originally developed with the purpose to support
the discrimination of different land covers (Angiuli et al., 2014). Additionally, two ratio bands dividing
shortwave infrared 1 (SWIR1) and near-infrared (NIR), and red and SWIR1 were included as suggested by
Saah et al. (2019).
Table 3. Band combinations for normalised differences; adapted from Saah et al. (2019).
Band 1 blue green red NIR SWIR1
Band 2 green red SWIR1 red SWIR2
Band 2 red NIR SWIR2 SWIR1
Band 2 NIR SWIR1 SWIR2
Band 2 SWIR1 SWIR2
Band 2 SWIR2
Furthermore, also the Enhanced Vegetation Index (EVI) (Jiang et al., 2008), the Soil-Adjusted Vegetation
Index (SAVI) (Huete, 1988), the Index-based Built-up Index (IBI) (Xu, 2008), the Enhanced Built-up and
Bareness Index (EBBI) (As-syakur, 2012), and the Tassel Cap transformation (Powell et al., 2009) were
included as covariate layers.
EVI has been successfully used in land cover studies, for example, by Hussein et al. (2017) and Wardlow et
al. (2007). Improvements in vegetation parameter sensitivity (e.g., Leaf Area Index (LAI)) in high biomass
regions are obtained through correction of the soil background below the canopy and a reduction in
atmosphere influences (Rayleigh scattering and ozone absorption) (Eq. 5) (Jiang et al., 2008).
EVI = G ∗NIR − red
NIR + C1 ∗ red − C2 ∗ blue + L (5)
Where G is a gain factor (2.5); C1 (6) and C2 (7.5) are the coefficients of the aerosol resistance term, and L
(1) is the soil-adjustment factor.
SAVI represents an advancement of the common NDVI which has limitations related to soil background
brightness, according to Bausch (1993). By including a soil-adjustment factor (L =0.5), different from the
one used in EVI, SAVI was found to adequately minimize soil background noise (Bausch, 1993). Therefore,
SAVI is superior compared to NDVI in detecting vegetation in areas with low plant cover (Xu, 2008). This
makes SAVI interesting with respect to the classification of young orchard plantations where plant cover is
still developing or for other sparsely vegetated areas where intensified soil background noise can be
expected. Equation 6 (Huete, 1988) was used for computation.
SAVI = (1 + L) ∗NIR − red
NIR + red + L (6)
Urban areas are complex systems which lead easily to spectral confusion. By using thematic Index-derived
bands rather than original images bands to construct the index, IBI suppresses background noise while
retaining urban features. The index is computed by considering the three major urban components, namely
soil, water and structures (Xu, 2008). This is achieved by making use of SAVI (Huete, 1988), the Modified
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
20
Normalised Difference Water Index (MNDWI) (Xu, 2006), and the Normalised Difference Built-up Index
(NDBI) (Zha, 2003). After Xu (2008) the IBI is defined by equation 7.
IBI =NIR
(NIR + red)+
green
(green + SWIR1) (7)
EBBI makes use of the thermal properties of urban areas to distinguish between built-up features and bare
land (As-syakur, 2012). Hence, the TIR of Landsat is included (Eq. 8) (As-syakur, 2012). The index was
considered as a useful covariate layer as its incorporation was expected to improve the classification in areas
where agricultural land is adjacent to urban areas.
EBBI =SWIR1 − NIR
10 ∗ √SWIR1 + TIR (8)
The Tassel Cap transformation derivatives were additionally used as covariate layers as they were reported
to be key variables for forest cover monitoring and mapping using RF by Boonprong et al., (2018). With
the coefficients from Crist and Cicone (1984) and the composite bands the Tassel Cap transformation
derivatives, “brightness”, “greenness”, ”fourth”, ”fifth”, and ”sixth” were calculated. Computations of the
Tassel Cap angle (tcAngle) and distances (tcDist) for all pairs of brightness, greenness and wetness were
done by equations 9 and 10, respectively (Powell et al., 2009).
tcAngle(band1, band2) = arctan (band1
band2) (9)
SRTM derivates
Terrain indices were derived from the SRTM digital elevation dataset. Computed properties included
elevation, slope, and aspect. Aspect was converted into two derivatives ‘eastness’ and ‘northness’. Eastness
is a measure of the deviation from east and was calculated by the sine of the aspect, whereas the northness
indicates the deviation from north, computed by the cosine of the aspect (Beers et al., 1966). The terrain
indices were used as covariate layers since they have proved to be valuable environmental variables for land
cover classification as shown by other studies, e.g. Brovkina et al. (2017), Domaç et al. (2006) and Koppad
and Janagoudar (2013). Furthermore, also conceptually a relation between the indices and the parameter to
predict (land cover) could be established:
• Elevation is expected to play a role for the land cover prediction, as cropland, for instance, is
cultivated in lowland areas where irrigation is easier, whereas mixed forest predominantly grows at
higher altitudes on summits, which are hard to access and therefore were spared by deforestation.
• Slope is expected to enhance the prediction of the land cover classes, as orchards were
predominantly observed on sloping areas, and cropland, as well as urban areas, could be found in
flat areas.
• Aspect is expected to influence the microclimate, including the amount of sunlight received, the
temperature achievable, and the moisture retained. All those parameters might influence vegetation
growth and therefore differences in land cover.
tcDist(band1, band2) = √band12 + band2
2 (10)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
21
3.3.5. Reference data
The RF model was trained with reference data which were collected for 2005 and 2018 (Table 4). The
reference data for 2018 were collected using a combination of high-resolution imagery (from 2018) available
in GEE and visual image interpretation of the different seasonal Landsat composites. Additionally,
OpenStreetMap (OSM) vector data (water bodies and road layer) were partially included as reference data.
For 2005, mainly the Landsat composite of 2005 was interpreted. Moreover, historical imagery in Google
Earth Pro served as a sampling basis, and the SERVIR-RCLMS land cover map (2005) was used in a
supporting manner. A greater amount of training points was collected for 2005 compared to 2018 since a
larger area (both watersheds) had to be classified and OSM data could not be used for 2005 (Table 4).
Table 4. Number of training data points collected in each class.
Land cover class No. training points
(2005) No. training points
(2018)
Orchard 819 262
Cropland 385 235
Teak Plantation 210 179
Mixed Forest 711 365
Urban 446 412
Water Bodies 103 -
Total 2674 1453
3.3.6. Machine learning
The number of potential covariates is large and, in some cases, provides redundant information, for instance,
when using the seasonal composites (e.g., SRTM derivatives). Eliminating covariates that provide little gains
in accuracy reduces computational resources and avoids the inclusion of noise. Therefore, at each reference
data point, all incorporated covariates, and each of the composites were sampled and the corresponding
values evaluated in R (Breiman, 2001; Liaw & Wiener, 2002; R Core Team, 2018). In the process, a principal
component analysis (PCA) was run on two measures which are commonly used to assess variable
importance, Mean Decrease Accuracy and Mean Decrease Gini (Saah et al., 2019). In the former instance,
the importance of a variable is assessed based on the mean decrease of the accuracy when that variable is
excluded. The more the accuracy of the RF decreases, the greater is the importance of that respective
variable for classifying the data. The latter measures the contribution of a variable to the homogeneity of
the nodes. A higher decrease in the Gini coefficient indicates that the respective variable results in nodes
with higher purity. Finally, all variables which resulted in a PCA coefficient greater than 0 were selected for
the classification, as they can be considered having relative importance for the RF classifier according to
Saah et al. (2019). Selected variables for both classifications can be found in Annex 3.7. Once the covariates
were put in place, the RF classifier in GEE was applied with the selected bands. The model was run with
100 trees as suggested by Saah et al. (2019), and the number of variables was set to the square root of the
number of variables (default of GEE). Not all reference points were included from the beginning. The
classification procedure was done on a step by step basis with many runs. Meaning, some reference points
were collected for each class and only then the RF model was applied. In a next step the resulting land cover
map was then used for additional sampling. In this way the reference dataset was built and sample points
leading to miss-classification could be excluded in the next run and replaced by others.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
22
3.3.7. Validation data
For 2018, 315 validation data points were composed out of different sources. Firstly, 58 out of 62 land cover
samples collected in the field were used as validation data points (Table 5), 6 sample points had to be
excluded because the respective land cover area was less than the pixel size (30x30 m) and therefore
considered as not suitable.
Table 5. Number of field validation points for each class.
Land Cover No. of samples No. of samples excluded
Teak Plantation 7 4
Orchard 40 0
Cropland 9 0
Mixed Forest 6 0
Total 62 4
Secondly, 1614 in GEE collected training points got a random number between 0 and 1 assigned (without
recurrence). Afterwards, they were divided into two sets, one including all points with values <0.1 (10 %)
and the other one containing all points with values >0.1 (90 %). The first set (<0.1) with 163 points was
exported as validation data, and the second set (>0.1) was used to train the model. Thirdly, 94 additional
validation points were collected using Google Earth Pro. For 2005, the 461 validation points were
exclusively collected in Google Earth Pro, making use of historical imagery. Table 6 shows the distribution
of the validation points among the various classes for both years.
Table 6. Validation data points for 2005 and 2018.
Land cover class No. validation points (2005) No. validation points (2018)
Orchard 100 71
Cropland 92 40
Teak Plantation 40 66
Mixed Forest 97 49
Urban 97 89
Water Bodies 35 0
Total 461 315
3.3.8. Accuracy assessment
In the past, the necessity of validation of image classification products was not widespread, and results were
simply considered to be correct (Congalton, 1991). With the rise of digital image classification, the
determination of the information value of classifications is now vital (Rwanga & Ndambuki, 2017). Hence,
creating an error matrix and computing the overall classification accuracy, which presents the simplest
measure to retrieve quality information, are now common practice for validation assessment in digital image
classification. One who wants more detailed information can make use of the user’s and producer’s accuracy.
In this study, the overall accuracy and user’s as well as producer’s accuracies were determined. The overall
accuracy considers the correctly classified reference points, which can be found on the diagonal of the error
matrix and divides it by the total number of references sites, providing an accuracy value in percent
(Congalton, 1991). The producer’s accuracy, on the other hand, quantifies how often real features on the
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
23
ground are correctly shown on the classification product. Hence, it is computed by dividing the number of
correctly classified reference points by the total number of references points of the respective class.
Accordingly, the user’s accuracy does indicate the reliability of the map by providing information on how
often a pixel on the map represents that class on the ground. Computation is done by the division of the
classified pixels with the row totals (Congalton, 1991).
Another discrete multivariate statistical technique used in this research is KHAT. Cohen (1960) introduced
this method where the resulting KAPPA coefficient is another measure of accuracy. Its upper limit is 1,
indicating perfect agreement, whereas results <0 are considered as poor. Detailed descriptions of all
mentioned methods can be found in Congalton (1991) and Rwanga et al. (2017). With equation 11, the
KAPPA coefficient was computed.
K=N ∑ xii−∑ (xi+ × x+i)r
i=1ri=1
N2−∑ (xi+ × x+i)ri=1
(11)
Where N = total number of observations, r = number of rows in the error matrix, xii = number of
observations in row i, xi and x+i = marginal totals of row i and column i, respectively.
3.3.9. Land cover change analysis
Various methods of remote sensing based land cover change detection are available, in-depth descriptions
can be found among others in Comber et al. (2016) and Mas (1999). For this study, post-classification change
detection comparison was used to identify and quantify land cover changes between 2005 and 2018. This
method was chosen as it represents a solid commonly used method characterized by its simplicity and the
capability of comparing satellite imagery captured under different conditions and with different sensors.
Examples of successful applications can be found in Kim (2016), Mallupattu and Reddy (2013) and Lubis
and Nakagoshi (2011) as well as in Mas (1999). In post-classification change analysis, two or more
classifications based on images from different times are examined in a comparative manner as already
suggested by the name. Due to the nature of this method, it is evident that the accuracy of the results highly
depends on the accuracy of the classifications used.
The ultimate goal of the land cover change analysis was to learn more about the dynamics in Ban Da Na
Kham watershed, in other words, to acquire information about the “from-to” processes between the land
cover categories. To fulfil this task, a change detection matrix was established. From such a matrix, the
magnitude and direction of the change can be extracted. According to Lubis et al. (2011), the diagonal values
in the matrix indicate areas which remained unchanged in each land cover class while off-diagonal elements
provide information about occurred changes. The off-diagonal elements contain information about area
increase and area loss of the different classes. Vertical values (columns) indicate an increase for a certain
class, whereby horizontal values (rows) indicate a loss of area in a certain class.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
24
3.4. OpenLisem flood model
Flash flood modelling was done in openLISEM to satisfy the objectives of this research and to answer the
following research questions:
Objective 1: Comparative analysis of SG versus FD for flash flood modelling;
• Is model calibration based on historical flood marks from a nearby watershed possible?
• What are the quantitative differences of the model output using, (I) FD, and (II) SG in relation to
flow dynamics?
• What is the sensitivity of the flood dynamics to different soil depths using, (I) FD, and (II) SG?
Objective 2: Analysis of the effects of land cover change on flash flood behaviour;
• Which land covers generate the highest average runoff?
• How do these land cover changes affect the runoff generation and flood dynamics?
OpenLISEM was chosen for this research study as it is an event-based integrated catchment model
operating in time steps of minutes and is therefore tailored to model flash flood processes. Furthermore, it
has the capability to account for detailed spatial variability in terms of land cover, soil and terrain. Moreover,
processes such as groundwater flow and evapotranspiration, which are superfluous when modelling single
events, are neglected (Bout et al., 2018). The applicability of openLISEM in flash flood modelling was
demonstrated by previous studies conducted in different parts of the world, e.g. by Pérez-Molina et al. (2017)
and Sliuzas et al. (2013) in Kampala, Uganda and Van Westen et al. (2015) in the Caribbean.
Originally the model was developed to estimate soil erosion but is now also used to assess flash floods and
debris flows. Modelled processes include, amongst others interception, infiltration, runoff generation, flow,
and sediment transportation. Vertical water flow in the soil, overland flow and channel flow, as well as
channel flooding, are important components of flash floods. The erosion component of the model includes
splash and flow detachment, as well as sediment transport and deposition (De Roo et al., 1996). Within
openLISEM land cover types are represented by various resistance values (Manning’s n) and hence influence
the flow behaviour. The effects of soils are determined by their respective infiltration rate and infiltration
capacity (porosity). Additionally, the model can include built structures such as urban areas or road networks,
which represent mostly areas where no or limited infiltration takes place.
The infiltration process is very important when modelling flash floods as it determines the amount of runoff
generated. In openLISEM infiltration is simulated by using the model of Green and Ampt (1911) which
assumes a downward moving wetting front due to surface water infiltration. Soil layers above the wetting
front are assumed to be saturated, and below the front, the soil possesses initial soil moisture. With equation
12, the potential infiltration rate is calculated and subsequently subtracted from the available surface water
(Bout et al., 2018).
fpot = −Ks (ψθs − θi
F+ 1) (12)
Where fpot = potential infiltration rate (m s-1), F = cumulative infiltrated water (m), θs = porosity (m3 m-3),
θi = initial soil moisture content (m3 m-3), ψ = matric pressure at the wetting front (m)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
25
Water that cannot infiltrate into the soil due to exceedance of infiltration rate or infiltration capacity and is
moreover not stored in micro depressions is described as runoff water. This type of water can result from
direct rainfall or from runoff generated on neighbouring cells. Furthermore, a Green and Ampt solution for
two soil layers is used with their characteristics; these are named soil layer SL1 and SL2 (see section below
on Soil Data).
Overland flow is simulated by either kinematic flow, diffusive flow or dynamic flow (Saint-Venant) (Bout
et al., 2018). When using kinematic flow, the water is routed via a predefined converging flow network.
Velocity is determined by friction forces and gravitational force. In the diffusive flow approximation,
velocity is additionally controlled by the hydraulic gradient. However, for this study, the dynamic flow was
used as it additionally takes the momentum terms into consideration. Furthermore, Bout et al. (2017) proved
that using the dynamic flow most accurate flow simulations could be achieved. According to Bout et al.
(2017), channel flow is generally decoupled from overland flow and routed by the kinematic wave. Whenever
runoff water reaches the channel it is added to the channel water. Flow in the channel is always computed
using a kinematic wave. In case of a channel overflow, this water is added to the overland flow water (Bout
et al., 2017).
3.4.1. Data preparation
OpenLISEM requires input data generated from four main data categories, namely rainfall, soil, land surface
and terrain properties (Table 7). To develop a modelling environment as close as possible to field conditions,
the parametrisation is done by including FD whenever available. By nature, this is limited due to time and
other resources. Hence, remote sensing products like a DEM, satellite images, as well as pre-existing GIS-
data and literature values support the input generation. Below the methodology used to generate the
subordinated input data for each category is elaborated.
Table 7. Main input data required for openLISEM.
Category Input data Comments
Rainfall Text file • Single rainfall event based on daily rainfall from 22nd of May 2006
To
po
grap
hy DEM -
Slope map • DEM derived
Local drain direction map (LDD) • DEM derived
Channels • Channels were generated based on LDD map; dimensions based on field measurements and Google Earth Pro
Surf
ace
Land cover • Generated land cover maps for 2005 and 2018
Manning’s n • Based on literature
Roughness • Based on literature
Vegetation cover • Based on field measurements
Leaf Area Index • Based on empirical equation
Road network • Generated based on OpenStreetMap data
So
il
Saturated hydraulic conductivity • Laboratory analysis and SoilGrids
Porosity • Laboratory analysis and SoilGrids
Initial soil moisture • Initial soil moisture based on antecedent rainfall and literature
Matric suction • Matric suction calculated based on initial soil moisture and soil texture
Depth • Based on topographical factors
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
26
Rainfall Data
Daily rainfall data for the period from 1951 to 2018, provided by the Thai Meteorological Department for
the meteorological station in Uttaradit city (Station ID: 351002), was processed. Based on Gumble analysis,
the return period (RP) of the extreme event in 2006 was determined. Subsequently, Intensity Duration
Frequency (IDF) curves for Uttaradit province published by Rittima et al. (2013) were used to generate 10-
minute rainfall intensities (Annex 4). Lastly, to compile a design storm, the Alternating Block method as
introduced by Yen and Chow (1980) was utilised.
Digital elevation model
The ALOS Palsar DEM with a spatial resolution of 12.5 m was obtained from the Alaska Satellite Facility
(ASF). With the tool gdalwarp the DEM was resampled to 15 m. Resampling with bilinear interpolation was
done for two reasons, (I) to speed up the modelling process but retaining sufficient spatial detail at the same
time, and (II) to smoothen the DEM. Smoothening represented a positive side effect of the resampling, as
the original DEM exhibited a “LEGO-like” surface due to the fact that elevation values are given as integers,
not as fractional numbers. Subsequently, a simple pit filling algorithm was applied to restore flow pathways.
Most of the ALOS Palsar radiometric terrain corrected products are created from SRTM DEM source,
which has a vertical accuracy of less than 16 m (ASF, 2015). Furthermore, C-band radar, which is used to
create these DEMs, does not penetrate well through areas of dense vegetation. Hence, these areas exhibit
elevations related to the top of the canopy (according to email correspondence with ASF).
For obvious reasons, it is expected that a DEM which does not relate to the bare soil surface might cause
considerable problems in hydrological modelling. Hence, it was decided to run openLISEM in the
calibration watershed with three different versions of the ALOS Palsar DEM, and additionally with the
freely available SRTM DEM from USGS. This choice was made to investigate differences in the model
output and to test the applicability of the DEMs for hydrological modelling in the watershed. The three
versions were assembled as follows: (I) original DEM (DEMo), (II) DEM without vegetation (DEMv), and
(III) DEM partially manipulated (DEMm). Generation of DEMv, was accomplished by subtraction of on-
site measured vegetation heights from DEMo. The average vegetation heights used can be found in Table
8. For DEMm elevation values were partially manipulated in narrow valleys along with the drainage network
in order to restore a functional drainage behaviour of the terrain as vegetation is expected to generate
artificial dams blocking the water flow.
Channels
Channel layers were created for both watersheds, for Ban Da Na Kham and for Laplae, based on the DEMo
derived LDD network. Validation of the channel location was subsequently done based on Google Earth
Pro. In the case of deviating channel locations caused by errors in the DEM, channels were manually
digitised in ArcGIS and PCRaster. This was mainly necessary for the lowland areas where rivers flow in flat
terrain. Besides channel location, also channel dimensions are an important input for openLISEM.
Therefore, dimensions were observed and measured during fieldwork at several locations in order to have
benchmarks for the channel parametrisation (Annex 5 & Annex 5.1). The channel dimensions for Laplae
watershed were approximated based on visual field observations and measurements in Google Earth Pro.
Surface parameters
A diverse set of surface parameters is necessary to represent the different land cover types in the modelling
environment (Bout et al., 2018). Manning’s n and surface roughness (RR) were parametrized based on
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
27
literature values. Vegetation overstory density (cover) was measured in the field using a spherical
densiometer (Model A), following the instructions of Lemmon (2008). Additionally, the average vegetation
height per land cover class was estimated based on field observations. To estimate the storage capacity of
the canopy, the LAI was calculated accordingly with equation 13. In Table 8, all surface parameters used are
shown.
LAI =ln (1 − cover)
−0.4 (13)
Soil data
Two sets of soil data were compiled, (I) FD analysed in the laboratory, and (II) SG data. Soil properties are
changing with increasing depth, as for instance SOM content and microbial activity are declining (Hassler,
2013). To be able to represent the soil landscape as realistic as possible, it was decided to use a two-layer
soil system. As no opposing information available, it was decided to close the soil at the bottom and
therewith prevent free drainage due to percolation.
Setup - FD:
For the surface layer (SL1) which possesses a thickness of 5 cm, Ks and porosity were parameterised based
on values derived from the soil analysis (Section 3.2.6). This value for thickness was chosen as it corresponds
to the thickness of the collected soil samples and additionally matches the distribution of the observable
SOM content and the presence of bioturbation. Only surface samples were collected. Therefore, no
immediate values for Ks and porosity for a second soil layer (SL2) were available. This lack was remedied
by using the PTFs by Saxton and Rawls (2006). As input for the PTFs, the PSD retrieved from laboratory
analysis was used. Since SHPs of deeper soil layers are not expected to be influenced by the land cover and
land use but rather by texture, it was decided to generate SL2 based on the aggregated PSD within each
slope unit and to set SOM to 0 %. Since the soil depth of SL2 is expected to vary in space, its generation is
discussed separately below.
Setup - SG:
For the SG setup PSD, Db, gravels and soil organic carbon (SOC) (g kg-1) layers for depths of 5 cm (SL1)
and 100 cm (SL2) were converted into Ks, porosity and matric suction maps based on the PTFs from Saxton
and Rawls (2006).
Initial Soil Moisture and Matric Suction:
Firstly, analysis of the historical daily rainfall data revealed that almost 135 mm of rainfall occurred in the
two preceding days of the event to model (Table 9).
Table 8. Surface parameter for openLISEM.
Land Cover Manning’s n Source RR* Cover (%) Height (m) LAI
Orchard 0.025 Morgan et al. (1998) 1 57 9.7 2.1
Cropland 0.04 Chow (1959) 0.1 47 0.9 1.6
Teak Plantation 0.1 Arcement and Schneider (1989) 1 80 21.5 4.0
Mixed Forest 0.2 Hessel et al. (2003) 1 78 14.3 3.8
Urban 0.0678 Kalyanapu et al. (2009) 0.5 2 9.7 0.6
Water 0 - 0.1 0 0.0 0.0
Channel 0.035 Chow (1959) - - - -
* RR adapted from Bout et al. (2018)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
28
Secondly, a post-disaster assessment report of the event in 2006 suggested that the tremendous landslides
and flash floods were caused by a saturation of the topsoil layer (30-40 cm) (Usamah & Arambepola, 2013).
Based on this information, the initial soil moisture content was set to 0.90 of the porosity to ensure
saturation during the rainfall event. However, this parametrisation introduces an unavoidable wrong
assumption that the soils are homogeneously saturated regardless of other influencing factors such as Ks,
land cover and soil depth. OpenLISEM only allows up to two soil layers when using Green and Ampt.
Furthermore, a gradual decrease of initial soil moisture corresponding to increasing depth can only be
implemented when the SWATRE module is used. This model is based on the full Richards equation for
multiple layers and beyond the scope of this research.
Subsequently, the matric suction (Psi) was calculated accordingly based on the Brooks and Corey Model
(Brooks & Corey, 1964) ( Eq. 14).
Psi = 𝑆𝑒(−𝜆)ℎ𝑏 (14)
Where 𝑆𝑒 = effective saturation, ℎ𝑏 = bubbling or air entry pressure head, and 𝜆 = is the pore size
distribution index. Values for ℎ𝑏 and 𝜆 depend on soil texture and were adapted from Saxton et al. (2006).
Psi maps for openLISEM were generated based on the slope unit map as soil texture is expected to relate
rather to the terrain than to the land cover. Correspondingly, for computation of Psi, the PSD of all soil
measurements within one slope unit were aggregated and an overall texture class estimated. Subsequently,
ℎ𝑏 and 𝜆 were extracted from Saxton et al. (2006).
Soil Depth:
Soil depth data was not available in the research area. During the fieldwork, soil depth measurements were
taken at the 48 soil sampling sites using a soil auger. These measurements indicated soil depths of at least 1
m at each location, but at the same time, the measurement activities were limited to 1 m because of the
length of the auger. Due to these limitations and having the knowledge that soil depth varies within the
landscape, the approach of Kuriakose et al. (2009) was adopted to generate a soil depth map. This approach
takes topographic factors such as altitude, slope, wetness index and profile curvature into account to
approximate soil depths within the landscape. A sort of calibration was done based on the field
measurements, meaning that at each measurement point the soil depth has at least 1 m depth. However, the
lack of a minimum and maximum soil depth, as well as the lack of validation data, led to a large unavoidable
uncertainty. In order to investigate this uncertainty, four additional soil maps were generated each having 1
additional meter of soil. In the course of the modelling, openLISEM was subsequently run with the SHPs
of FD and SG with varying soil depths to investigate how flow dynamics change in the watershed with an
increase in soil depth. The PCRaster script to generate the soil depth maps is available in Annex 6.
Table 9. Daily rainfall prior, during (red square) and after the event.
Date (May 2006) 14 15 16 17 18 19 20 21 22 23 24
Rainfall (mm) 25.6 17.4 0.5 0 7.9 0 14.5 120.4 263.7 13.4 8
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
29
3.4.2. Model calibration
After the flash flood event on the 22nd of May 2006 the community of Laplae painted marks on the power
pols along the major roads in the city of Si Phanommat to indicate the flood height and flood extent (Figure
8). Using a measuring tape and GPS receiver, data on the historical flood heights were captured in 45
locations (Figure 8). Based on these measurements, attempts were taken to calibrate the model.
For this purpose, the model was run in Laplae watershed using the three versions of the ALOS Palsar DEM
and the SRTM DEM. Subsequently, the number of points with a simulated flood depth ≥0.3 m was
calculated to identify the DEM having the best agreement in terms of the presence of flood. The water
depth of 0.3 m was selected as lower limit as every inundation below that threshold was expected to
represent the effects of pure precipitation rather than by channel flooding.
Figure 8. Locations of historical flood height measurements in Si Phanommat city (left) and flood mark painted on a power pole (right).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
30
4. RESULTS AND DISCUSSIONS
4.1. Soil analysis
In this subsection, the results of the general statistical analysis of soil data are described. Firstly, the results
of the slope unit delineation are presented, as it served as an input for further analysis. Secondly, the FD
was put in relation to land cover types and slope positions to investigate possible effects and how they are
reflected by the soil properties. Thirdly, an examination of SG based on the land covers and slope positions
was done to see if similar patterns compared to the FD can be observed. In the last part, similarity analysis
between the FD and SG is presented, making use of the Cosine Similarity measure and the Wilcoxon Signed-
Ranked test. In the course of this section, the following research questions will be dealt with:
• How well do SG and FD correlate?
• How do the soil properties of (I) FD, and (II) SG relate to the main land cover types?
• How do the soil properties of (I) FD, and (II) SG relate to the terrain?
4.1.1. Slope units
Slope unit delineation was aided by the GRASSGIS extension called “Geomorphon”. Best results were
achieved applying a search radius of 2000 m, a skip radius of 15 m, and a flatness threshold of 0. Figure 9
shows the spatial distribution of the slope units in the research area, with the narrow valley floors and
similarly narrow summits. The accuracy assessment of the slope position delineation yielded an overall
accuracy of 79 % (Table 10). Misclassification is expected to be caused by the deviation of the GPS location
due to poor signal within the forested and mountainous areas. Also, the resolution and quality of the DEM
need to be taken under consideration as a potential source for wrongly classified pixels. According to Miller
et al. (2015), the identification of small variations in topography and the correct classification of slope
positions does depend on the DEM quality and resolution, used for the classification. This seems to be
especially important in areas where valley floors and summits are located relatively close to each other, due
to steep sloping terrain, as observed in the research area. Where one pixel of the DEM represents the
elevation of the summit, an adjacent pixel could already provide information on the valley bottom.
Furthermore, should the DEM quality issues as described in Section 3.4.1 be taken into consideration.
Having the vegetation height on top, partly wrong classified pixels can be expected.
Figure 9. Slope position map.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
31
4.1.2. Influence of land cover on soil properties
Various descriptive statistics were computed for the FD (Table 11 & Figure 10). Striking are the field
measurements of Ks, with values ranging from 0.1 mm h-1 up to 10784 mm h -1. It is worth to note that Ks
values above 2000 mm h-1 were considered as arbitrary high and therefore excluded from further analysis.
Details on the omitted data can be found in Annex 2.3.
Table 11. Statistics of soil properties per land cover.
Land Cover Variable
n
Clay (%) Silt (%) Sand (%) Ks (mm/h) Por (%) Db (g/cm3) SOM (%)
OR
Mean 18 42 40 411 54 1.2 4.7 34
SD 11 7 14 446 6 0.2 1.3
CR
Mean 21 37 42 65 42 1.5 2.5 4
SD 7 9 15 23 6 0.2 0.4
TK
Mean 7 42 51 594 51 1.3 4.6 2
SD 7 5 2 177 7 0.2 2.0
MF
Mean 30 40 30 722 61 1.0 5.2 8
SD 10 6 14 750 8 0.2 1.8
OR = Orchard, CR = Cropland, TK = Teak Plantation, MF = Mixed Forest, Ks = Saturated Hydraulic Conductivity, Por = Porosity, Db = Bulk Density, SOM = Soil Organic Matter, SD = Standard Deviation, n = number of samples
In the research area, it becomes obvious that mixed forest represents the land cover class with the highest
max Ks values (1986 mm h-1), followed by perennial crops comprising long kong trees and banana plants
with 1782 mm h-1, and teak plantations with 719 mm h-1 (Figure 10). Both orchards and mixed forest show
also a notable high standard deviations with 446 mm h -1, and 750 mm h -1, respectively, indicating great
variation within the investigated territory (Table 11). The exceptional high Ks values measured might be
explained by the intensive bioturbation in the form of abandoned root channels and holes excavated by
insects living in the near-surface layer as well as high SOM content that were observed during the fieldwork.
This observed near-surface soil layer condition is additionally reflected by high porosity values of up to 76
%, the corresponding Db of 0.6 g cm-3, and SOM content of on average more than 6 % (Figure 10 & Table
11). The findings are in line with observations made by Hassler et al. (2011) and Hassler (2013), who
investigated Ks variability in the tropics and reported measurements of more than 1400 mm h-1 within
forested areas. In contrast, Hassler (2013) reported the lowest measurements of 0 mm h -1 in all land cover
classes, referencing to great variability.
Table 10. Confusion matrix of slope unit classification.
Geomorphon Field Observations Row
Total
User Accuracy (%)
Backslope Summit Valley
Backslope 18 4 2 24 75
Summit 4 4 100
Valley 4 16 20 80
Column Total 22 8 18 48 Total Accuracy 79%
Producer Accuracy (%) 81.8 50 88.8
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
32
In order to get an overview of the data and to find possible linkages, Ks was correlated with ancillary variables
such as Db, SOM content and PSD. This was done using all data points based on the land cover or slope
position without any subdivision. Ks was kept as a constant variable during correlation because it is known
that other ancillary variables influence Ks (Hassler et al., 2013). Subsequently, the same procedure was
applied, but in each land cover class (Table 12).
Since the macro-porosity of soils has a great impact on Db, which again has a positive relationship with Ks
(Ahuja et al., 2010), it can be argued that drivers affecting Ks similarly affect Db. Examples for such drivers
are abandoned root channels or microorganism activity. Furthermore, SOM content has a positive
relationship with soil structure (Lado et al., 2004). Thus, it is expected that Ks shows an increase with
increasing SOM content. Also, the percentage of sand, silt and clay may influence the Ks, whereas after
García-Gutiérrez et al. (2018) the main determinant in respect to Ks is the percentage of sand.
A statistically significant correlation with P≤0.01 and a medium correlation coefficient of r=0.5 was found
between Ks and Db considering the complete dataset. At the same time, correlations with other variables
were rather low and not statistically significant.
Further, it is unlikely that the texture class influences the Ks of the collected soil samples indicated by the
absence of correlation between the percentage of sand, silt and clay. For the tropics, Hassler et al. (2011)
compared spatial means of Ks between different soil map units and did not observe any differences.
Figure 10. Variability of soil properties per land cover; a) Saturated Hydraulic Conductivity, b) Soil Organic Matter, c) Porosity, d) Bulk Density.
a) b)
c) d)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
33
Likewise, Sobieraj et al. (2002) faced difficulties with explaining changes in Ks along a catena and Hassler
(2013) also failed to identify a correlation between differences of Ks and sand, silt or clay content. In line
and with the thoughts of the above-mentioned authors, it can be concluded, that it is very likely that biotic
factors predominate the formation of soil properties within the study area.
In the next step, Ks measurements were correlated with the ancillary variables within different land cover
classes. Since the available Ks measurements for cropland was reduced to 3 after applying the threshold
mentioned above, and the teak plantation class was just represented by two samples, the correlation was
done within the orchard and mixed forest class only (Table 12). Subsequently, a statistically significant result
were found within the mixed forest class with P≤0.05 and a high correlation of r= -0.92 with Db (Table 12).
Additionally, SOM and Ks correlated very well with r=0.75, but no statistical significance was exhibited
(Table 12). This confirms the observations from Lado et al. (2004) that SOM positively influences the soil
structure and therewith Ks, and at the same time diminishes compacted soil conditions.
More than 55 % of the analysed orchards are located in steep sloping terrains, which favour the loss of
surface material due to erosional processes (Schaetzl et al., 2005), even in tropical forest-like ecosystems
(Sidle et al., 2006). Thus, it can be assumed that the considerably high variability expressed by the standard
deviation of each orchard class and the significant relationship with P≤0.05 and r=0.37 between Ks and clay
particles originates partly from erosion (Table 12 & Table 13). Loss of the litter cover and an increase of
soil surface exposed to precipitation could lead to a poorly developed A horizon and thereby potentially
transform an alleged surface sample in a sub soil sample according to Hassler (2013). This result urges the
possibility that the influence of land cover is decreasing with increasing soil depth and that soil properties
of deeper soil layers, e.g. below 5 cm depth, might rather be controlled by soil texture.
Interesting is the fact that in the orchard class, just weak correlations were found between Ks and SOM and
without any statistical significance (Table 12). This raises the question of why a strong relationship was
observed within the mixed forest class (Table 12). Nemes et al. (2005) discussed similar contradictory
interactions of SOM and soil properties. An explanation could be a possible enhancement of the flow of
water through the soil in the mixed forest class by improved porosity. On contrary to that, the SOM in the
orchard class could cause a filling of pores in the soil, leading to tortuous and thin pathways decelerating
the water flow. This would result rather in water retention by SOM, explaining the weak correlation of SOM
and Ks within the orchard class. Another reason could be that the floor of the mixed forest was covered by
small seedlings and dense undergrowth which are potentially also a source of soil structure enhancement
next to the SOM. In conclusion, SOM content seems to have a varying effect on the soil water characteristics
within different land cover classes in the study area and even within just one land cover class.
In order to gain a better understanding of the variability within the orchard class, the class was divided into
subclasses based on fruit trees; namely long kong, banana and long kong (mix), and pure banana.
Table 12. Correlation (r) of ancillary variables with Ks per land cover.
Variable Orchard LK Mix Banana Cropland TP Mixed Forest
Db (g/cm3) n.s. n.s. n.s. n.s. n.a. n.a. -0.92* Clay (%) 0.37* n.s. n.s. n.s. n.a. n.a. n.s. Silt (%) n.s. n.s. n.s. n.s. n.a. n.a. n.s. Sand (%) n.s. n.s. n.s. n.s. n.a. n.a. n.s. SOM (%) n.s. n.s. n.s. n.s. n.a. n.a. 0.75 Db = Bulk Density, SOM = Soil Organic Matter, LK = Long Kong, Mix = Banana and Long Kong, TP = Teak Plantation, n.s. = not significant, n.a. = not available, * = statistically significant at P<0.05 level
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
34
The disaggregation held interesting insights as the Ks, Db and SOM differ within the subclasses (Table 13).
It can be observed that mean Ks and mean SOM content decrease from pure long kong plantation with Ks
of 521 mm h-1 and SOM of 4.9 % to areas with only banana trees planted showing a Ks of 140 mm h-1 and
a SOM of 4.1 % (Table 13). In contrast to that, the mean Db is increasing from 1.1 g cm-3 to 1.3 g cm-3
indicating increasing compaction (Table 13). Also, the composition of sand, silt and clay is going through a
transition with having a higher mean percentage of clay (23 %) within the pure long kong plantations
decreasing to a mean of 11 % in the pure banana plantations. At the same time, the sand component is
rising more than 16 % to a mean of 51 % (Table 13). The finding entails that soil texture is not necessarily
the predominant factor causing the alteration of the soil physical properties within the fruit tree classes.
Rather it can be assumed that it is the source and proportion of organic material determining the changes
in the physical properties. Hence, overripe fruits dropped by the long kong trees, as observed during the
fieldwork could be one cause for the differences in SOM content among the orchard types. This
phenomenon was investigated by several studies, e.g. see De Paz et al. (2018), Ohm et al. (2007) and Hamer
and Marschner (2005). Leaking liquid from the fallen fruits enriches the soil with sugar, which intensifies
the microbial growth and consequently enhances the decomposition rate of the fallen leaves. Further, an
acceleration of the leaf decomposition was observed by De Paz et al. (2018), not only triggered by the
absolute sugar content of the fruits, rather by factors such as fruit moisture and micronutrient content. This
corresponds with the properties of long kong fruits being fleshy with approximately 84 g water per 100 g
long kong (Tilaar et al., 2008).
Table 13. Soil properties per orchard subclass.
Land Cover Variable n
Clay (%) Silt (%) Sand (%) Ks (mm/h) Por (%) Db (g/cm3) SOM (%)
LK
Mean 23 45 32 521 57 1.1 4.9 34
SD 10 5.5 12.8 409 4.2 0.1 1.3
Mix
Mean 16 41 43 470 55 1.2 4.6 4
SD 8.6 6.4 10.4 581 5.4 0.1 1.5
BN
Mean 11 38 51 140 49 1.3 4.1 2
SD 10.1 9.2 11 134 6.2 0.2 1.0
LK = Long Kong, Mix = Banana and Long Kong, BN = Banana, n = number of samples, Ks = Saturated Hydraulic Conductivity, Por = Porosity, Db = Bulk Density, SOM = Soil Organic Matter, SD = Standard Deviation
In contrast, Ks in the cropland is moderate, with an average value of 65 mm h-1 and a standard deviation of
23 mm h-1 (Table 11). Comparable high is Db with a mean of 1.5 g cm-3 and a maximum of 1.7 g cm-3
corresponding to a relatively low SOM content with a mean of 2.5 % and a maximum of 2.9 % (Table 11
& Figure 10). Increased Db can also be an indication of disrupted pores by cultivation, as outlined by
Kizilkaya and Dengiz (2010).
It should be noted that predominantly areas with corn were sampled while most of the cropland in the
research area is represented by paddy fields. Unfortunately, paddy fields could not be sampled, being already
tilled and flooded. A drastic drop of mean Ks within the cropland class is expected if the soils of paddy fields
would be taken into account. This expectation is supported by Wijaya et al. (2009) and Boumana et al. (1994)
who investigated Ks within flooded paddy fields. Their results show values ranging from 0.0125 mm h-1 up
to a maximum of approximately 0.208 mm h-1 for poorly permeable puddled plow soles.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
35
4.1.3. Field data related to the terrain
In order to identify possible linkages between slope positions and soil properties descriptive statistics such
as mean and standard deviation were computed for the classes summit, backslope and valley, and box plots
were created (Table 14 & Figure 11). Additionally, soil properties were correlated against slope gradient and
elevation, and Ks with ancillary variables such as Db, porosity and SOM (Table 15 & Table 16).
The correlation with the slope gradient did not yield any significant results. Whereas elevation showed a
moderate (r=0.45) but significant (P≤0.01) correlation with clay content signifying the occurrence of higher
clay content in higher altitudes (Table 15). This can also be observed in Table 14 and Figure 11, where
summits represent with a mean of 23 % and a max of 42 % the slope position with the highest clay content.
At the same time, sand content seems to increase in lower terrain as indicated by a moderate but significant
correlation with P≤0.05 and r=-0.3 (Table 15) and the highest mean sand content being 43 % in valleys
(Table 14).
Clay particles are difficult to detach, and their cohesion can be enhanced by organic matter rich soils (Gilley,
2005), like the once found on summits and backslopes (Table 14). Once clay particles are detached, they can
be transported easily by overland flow. On the contrary, sand particles generally lack cohesiveness and are
thus detached easier. Because of their size, sand particles require more energy to be transported (Gilley,
2005), hence, their greater occurrence might be related with an increase in flow accumulation in lower
altitudes of the watershed. Furthermore, it can be assumed that eroded silt and clay particles were already
transported out of the watershed and deposited in downstream areas. Apart from that, the occurrence of
sand, silt and clay varies not much among the different positions in the upper most layer. Slight differences
are detectable but not as much as that they could be linked to erosional and dispositional processes as
described by Malo et al. (1974).
By correlating Ks values with ancillary variables within each slope unit a significant relationship (P≤ 0.01;
r=-0.75) with Db was detected in the backslope class, as well as a strong dependency of Ks on SOM content
on summits (r=0.81) (Table 16). The latter can be traced back to the fact that >83 % of the samples collected
within the summit class were located within the land cover mixed forest and orchard. SOM’s dominating
impact on soil physical properties in those classes was discussed in the above section.
Table 14. Soil properties per slope unit.
Position Variable
n
Clay (%) Silt (%) Sand (%) Ks (mm/h) Por (%) Db (g/cm3) SOM (%)
SU
Mean 23 41 36 570 58 1.1 4.7
6 SD 13.8 10.4 20.6 599.8 7.6 0.2 1.3
BS Mean 20 43 37 434 56 1.2 4.9
24 SD 11.9 6.1 14.2 490.5 7.2 0.2 1.6
VA
Mean 18 39 43 398 51 1.3 4.0
18 SD 10.0 7.0 11.5 457.1 7.2 0.2 1.2
VA = Valley, BS = Backslope, SU = Summit, Ks = Saturated Hydraulic Conductivity, Por = Porosity, Db = Bulk Density, SOM = Soil Organic Matter, SD = Standard Deviation, n = number of samples
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
36
In terms of Db a slight increase of the mean and max values is inherent starting in the catena from the
summit (mean = 1.1 g cm-3; max = 1.4 g cm-3) via the backslope (mean = 1.2 g cm-3; max = 1.6 g cm-3) to
the valley (mean = 1.3 g cm-3; max = 1.7 g cm-3) (Table 14 & Figure 11). This can be attributed to a transition
of land covers along the catena and to compaction in agricultural areas in the valleys.
In summary, a more dominant impact of the different slope positions on the soil PSD and therefore, on the
soil physical properties, was expected. It is evident that the soil properties are not varying too much along
the catena. Especially the soil properties of the summit and backslope class are alike. When comparing soil
properties on slope units (Table 14 & Figure 11) with soil properties in different land cover types (Table 11,
Table 13 & Figure 10) it seems as if values become normalised by aggregation, and specifications attributable
to land covers get lost. The suspicion that land cover and land use, as well as biotic factors, are the prevailing
determinants of the soil properties in the upper soil layer as discussed earlier seems to be confirmed. This
result coincides with the findings of Hassler (2013), Hassler et al. (2011) and Sobieraj et al. (2002) who
discovered comparable conditions in tropical soils along slope transects.
Variable Elevation Slope
Db (g/cm3) n.s. n.s.
Clay (%) 0.45** n.s.
Silt (%) n.s. n.s.
Sand (%) -0.3* n.s.
SOM (%) n.s. n.s.
*/** = statistically significant at P≤0.05/P≤0.01 level, n.s. = not significant, Db = Bulk Density, SOM = Soil Organic Matter
Variable Summit Backslope Valley
Db (g/cm3) n.s. -0.75** n.s.
Clay (%) n.s. n.s. n.s.
Silt (%) n.s. n.s. n.s.
Sand (%) n.s. n.s. n.s.
SOM (%) 0.81 n.s. n.s.
** = statistically significant at P≤0.01 level, n.s.= not significant, Db = Bulk Density, SOM= Soil Organic Matter, Ks = Saturated Hydraulic Conductivity
Table 16. Correlation of Ks per slope unit with soil properties.
Table 15. Correlation of terrain derivates
with soil properties.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
37
Figure 11. Variability of soil properties per slope unit; a) clay, b) sand, c) silt, d) Saturated Hydraulic Conductivity, e) Porosity, f) Bulk Density, g) Organic Matter Content.
g)
a)
c) d)
e) f)
b)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
38
4.1.4. SoilGrids variability in the landscape
Descriptive statistics for all soil properties were calculated, for the ones originally extracted from the SG
layers and for the ones which were created by the PTFs of Saxton et al. (2006), such as Ks and porosity. The
data can be found in Annex 2.5. Statistics were computed for each type of land cover and the three slope
positions (Table 17 & Table 18). It is evident that spatial variability is not represented by the SG dataset,
neither in different land cover classes nor on the slope positions. The largest standard deviation was
computed for Ks with 8.9 mm h-1 and the lowest can be found for Db with 0 g cm-3 and porosity with 0 %,
which indicates a very low or not existing variability (Table 17). The same applies to the PSD, the percentage
of sand, silt and clay is almost identical in all of the three slope position (Table 18). This contradicts the
investigations of Malo et al. (1974) who observed different processes taking place along a catena, leading
spatially to different PSDs, causing differences in soil properties.
Significant are the continuously high Db values being lowest in the mixed forest on the summit with
1.3 g cm-3 and greatest in almost all land covers with 1.6 g cm-3 (Table 17 & Table 18). Those high values
seem to be generic and do not take into account the impact of land cover types and land use practices. Thus,
they do not represent the mostly porous and lose surface soil layer caused by abandoned root channels,
microorganism activity and other biotic factors as observed during the fieldwork. Further, they contradict
the findings of Hassler (2013) who reported for tropical forest soils Db of 0.8 g cm-3 and correspond more
to the findings of Muñoz et al. (2007) who reported Db of greater 1.5 g cm-3 for very degraded soils in Chile.
The mean SOM contents of 5-6.5 % having its highest values with 7 % and 8 % in the mixed forest and
orchard class respectively appears valid for a tropical area although a lack of spatial variability is recognised
(Table 17). This is identifiable by a mean SOM content of 6 % in agricultural areas, however, ignoring the
process of SOM declination by deforestation activities and continuous cultivation as it is described by Ross
(1993) (Table 17). SG makes use of different soil covariates to predict soil properties (Hengl et al., 2017).
One of these covariates is the land cover product GlobalCover30 of the year 2010 by Chen et al. (2015).
This could be another reason for missing spatial variability of SOM content, as this product is clearly
outdated, and land cover changes are therefore not taken into consideration.
It is evident that the PTFs by Saxton et al. (2006) considers SOM as soil structure enhancing factor.
Therefore, an increase in SOM will always result in greater porosity and Ks and a decline in Db. Because
different soils and different types of SOM might have different intercorrelation, this feature is questionable.
Where one type of SOM influences soil aggregation and associated pore space distribution (Hudson, 1994)
another one would cause clogging of available pores and thereby reduces the flow of water (Nemes et al.,
2005). Signs of this behaviour were, for example, visible during the assessment of the FD where Ks exhibited
a strong positive correlation with SOM in the mixed forest. In contrast, no similar correlation was found
within the orchard class.
Ks values produced by SG and the PTFs are largely controlled by the PSD, which is visible in Table 17.
Increasing clay by a few percent will result instantly in a decrease of Ks. Therefore, this combination might
potentially achieve good results when applied for estimating soil-water characteristics of deeper soil layers
where the influence of surface processes is not dominant, and texture takes over control.
Using the PTFs of Saxton et al. (2006) to predict soil properties such as porosity and Ks is convenient but
should be used with caution. The PTFs are based on the original PTFs introduced by Saxton et al. (1986)
and were considered as the best performing PTFs during an extensive review conducted by Gijsman et al.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
39
(2003). However, the output of a PTF is largely depended on the soil database used for its definition, and
thus, its predictions reflect the interactions of its input soils (Nemes et al., 2005).
Table 17. Statistics of SoilGrids soil properties per land cover.
Land Cover Variable n
Clay (%) Silt (%) Sand (%) Ks (mm/h) Por (%) Db (g/cm3) SOM (%)
Orc
har
d
Mean 26 32 42 36 59 1.5 6
34 SD 1.7 1.4 1.8 7.0 2.4 0.1 1
Min 22 29 37 17 53 1.4 4
Max 30 35 46 57 65 1.6 8
Cro
pla
nd Mean 27 32 41 34 59 1.5 6
4 SD 3.0 1.2 1.9 4.5 0.9 0.1 0.4
Min 23 32 39 31 58 1.4 5
Max 29 34 43 41 59 1.6 6
Tea
k
Pla
nta
tio
n Mean 30 32 38 22 56 1.5 5
2 SD 0.7 1.4 0 0.2 0 0 0.2
Min 29 31 38 22 56 1.5 5
Max 30 33 38 22 56 1.5 5
Mix
ed F
ore
st Mean 28 32 40 29 57 1.5 5
8 SD 2.6 1.3 1.5 7.0 2.0 0.1 0.9
Min 24 31 38 23 55 1.3 4.5
Max 30 35 41 41 61 1.6 7
Ks= Saturated Hydraulic Conductivity, Por = Porosity, Db = Bulk Density, SOM= Soil Organic Matter
Table 18. Statistics of SoilGrids soil properties per slope unit.
Position Variable n
Clay (%) Silt (%) Sand (%) Ks (mm/h) Por (%) Db (g/cm3) SOM (%)
Sum
mit
Mean 27 32 41 37 60 1.5 6.5
6 SD 2.0 1.2 1.0 4.7 1.7 0.1 0.9
Min 24 31 39 32 57 1.3 4.8
Max 29 34 42 41 61 1.6 7.2
Bac
ksl
op
e
Mean 27 32 41 33 58 1.5 5.8
24 SD 2.0 1.3 2.2 8.9 2.8 0.1 1.2
Min 24 29 37 17 53 1.4 3.6
Max 30 35 46 57 65 1.6 8.3
Val
ley
Mean 26 33 41 34 58 1.5 5.8
18 SD 2.2 1.5 1.8 6.0 2.0 0.1 0.9
Min 22 30 38 22 56 1.4 4.5
Max 30 35 44 42 62 1.6 7.4
Ks = Saturated Hydraulic Conductivity, Por = Porosity, Db = Bulk Density, SOM = Soil Organic Matter
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
40
4.1.5. Similarity analysis of SoilGrids and field data
In order to quantify the similarity of SG data and FD, the Cosine Similarity was used as similarity measure.
To make a statement if FD and SG are similar from a statistically perspective, the Wilcoxon Signed-Rank
test was conducted.
Since SG purest form is the PSD, the first assessment was done just comparing the percent of sand, silt and
clay of the two datasets. In this manner, it can be prevented to include an uncertainty caused by other
predicted soil physical properties through the use of PTFs. The Cosine Similarity for the summed PSD
values scored a similarity of 0.98, indicating a high level of similarity. In the next step, the same procedure
was done but considering different slope positions. The highest results were found in the valley and summit
class with a cosine of 0.99 and the lowest in the backslope class with 0.97 (Table 19). However, all of them
show high similarity within the fractions of sand, silt and clay.
In the next step, the other soil properties (normalised) such as Ks, Db and SOM were compared. This time
based on the different land cover types as it was found that land cover has a great impact on these properties
within the research area (Section 4.1.2). Table 20 shows that high similarity was yielded in each land cover
class, being lowest in the cropland class with 0.68. Additionally, different combinations of soil properties
were tested with the Cosine Similarity for each site. The results can be found in Annex 6. Overall, almost
always a high level of similarity was found. Each combination gave an average similarity of 0.77. The lowest
result for a single site was 0.42, which can still be considered as moderate similar (Annex 7).
Achieving high similarity throughout the datasets seems to be odd considering the different behaviours of
the datasets as discovered in the previous sections. On the one hand, having the FD which is highly variable
and on the other hand having SG showing almost no variability in the landscape (Section 4.1.2 & 4.1.4).
Hence, an obvious assumption is that the Cosine Similarity does not represent an appropriate similarity
measure when comparing soil data due to a lack of sensitivity. Differences in properties must be very high
to have a significant effect.
Before the Wilcoxon Signed-Rank test could be computed, both datasets had to be assessed in terms of
normality by deploying the Shapiro-Wilk test, kurtosis and skewness. A normal distribution is given if the
z-values (at P>0.05) of skewness and kurtosis fall within the span of -1.96 to +1.96 and if the Shapiro-Wilk
P-Value is above 0.05 (Field, 2009). Table 21 shows the results of the normality test, indicating that not all
property pairs were normally distributed. Examples for that are the Ks values (FD) and the silt and SOM
values (SG) with a Shapiro Wilk P-Value of 0, 0.001 and 0.037, respectively (Table 21). As not all properties
were normally distributed the Wilcoxon Signed-Rank test was seen as an appropriate choice.
Table 19. Cosine Similarity per position.
Position Cosine Similarity
Summit 0.99 Backslope 0.97 Valley 0.99
Computation was done based on the particle size distribution
Table 20. Cosine Similarity per land cover.
Land Cover Cosine Similarity
Orchard 0.94 Cropland 0.68 Teak Plantation 0.97 Mixed Forest 0.99
Computation was done based on the soil properties: Ks, Db and SOM
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
41
Table 21. Results of the test of normality.
Dataset Property Skew
Z-Value Kurtosis Z-Value
Shapiro Wilk P-Value
Fie
ld D
ata
Sand 0.43 -1.063 0.085
Silt -0.321 0.631 0.416
Clay 0.026 -0.718 0.203
Db 0.005 0.91 0.767
SOM 0.408 -0.132 0.33
Ks 4.817 27.45 0.000 So
ilGri
ds
Sand -0.041 0.059 0.253
Silt -0.46 -0.553 0.001
Clay 0.221 -0.615 0.101
Db -0.05 0.193 0.744
SOM -0.08 -0.867 0.037
Ks 0.154 0.855 0.151
Ks = Saturated Hydraulic Conductivity, Db = Bulk Density, SOM = Soil Organic Matter
Table 22 shows the results of the Wilcoxon Signed-Rank test. Sand was the only physical property which
yielded a P>0.05, declaring sand as the only property having statistical similarity when comparing FD and
SG. The other properties had P-Values consistently below the critical value of 0.05. Therefore, it can be
concluded that there is a significant difference inherent.
Positive and negative ranks indicate the direction of reported difference (Table 22). Hence, the majority of
negative ranks within Db, SOM and Clay indicate that the FD has significantly lower values compared to
SG, suggesting an overestimation by SG. In contrast, when inspecting Ks and Silt, it is evident that the FD
is significantly higher, indicating an underestimation of the properties by SG.
These results imply that SG data should be used carefully. Especially when used for hydrological modelling
on a watershed scale where differences in water holding capacity or infiltration are decisive. For instance,
clay content and Db are clearly overestimated by SG (Table 22). This, in turn, will greatly affect other soil
properties such as Ks. Having predominately an underestimation of Ks values predicted with the PTFs
(Table 22) the model will generate surface runoff when rainfall intensities exceed the infiltration rate
although infiltration would take place in reality. Similarly, when looking at SOM, an overestimation could
lead to a wrong water holding capacity of the soil. Thus, rainfall intensities below Ks might infiltrate in the
soil whereas a soil with less SOM could already be saturated and hence surface runoff would occur much
earlier.
Table 22. Output of Wilcoxon Signed-Rank test on SoilGrids derived properties.
Property Standard Error Z-Score P-Value Positive Ranks (n) Negative Ranks (n)
Sand 97.5 -1.1 0.272 21 27
Silt 97.5 5.44 0.000 43 5
Clay 97.5 -3.3 0.001 14 34
Db 97.5 -5.7 0.000 4 44
SOM 97.5 -4.3 0.000 13 35
Ks 97.5 5.6 0.000 41 7
Ks = Saturated Hydraulic Conductivity, Db = Bulk Density, SOM = Soil Organic Matter
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
42
4.1.6. Summary of soil analysis
Spatial variability and the effects of land cover were investigated for (I) FD and (II) SG. Subsequently, the
similarity of both datasets was assessed using Cosine Similarity and the Wilcoxon Signed-Rank test. The
result of the independent assessment of the FD revealed that the surface soil layer in the watershed and its
physical and chemical properties is highly affected by the respective land cover class and land use activities.
Those effects are so pronounced that even clear patterns between different types of fruit trees in the orchard
class could be established. Furthermore, SHPs such as Ks and porosity seem to be mainly determined by
the presence of intensive bioturbation. Soil texture, on the other hand, plays a subordinated role. In contrast
to that, the influence of the terrain and the different processes taking place is neglectable, as soil properties
do not vary much along the catena. The assessment of SG data revealed its almost non existing spatial
variability within the watershed. Neither effects of different land cover types nor of changing slope positions
are identifiable. Furthermore, PTFs should be used carefully as they reflect the interactions of soils used for
their creation and do not necessary represent features which are location specific.
The similarity analysis of the two soil datasets gave contradicting results. Cosine Similarity, on the one hand,
yielded a high similarity of both FD and SG, regarding soil properties within the three slope positions, and
different land covers. The Wilcoxon Signed-Rank test, on the other hand, declared sand as the only property
having a statistical similarity. Since obvious differences between both datasets were observed during the
independent assessment, Cosine Similarity is expected not to be a suitable similarity measure for soil
properties. The Wilcoxon Signed-Rank test furthermore indicated that fundamental soil properties such as
percentage clay and SOM, as well as Db, appeared to be underestimated by SG compared to field data which
had a great effect on SHPs predicted by the PTFs and potentially far-reaching consequences for the
subsequent hydrological modelling.
Based on the obtained findings, several conclusions for the model parametrization can be drawn. Firstly,
since the properties of the surface soil layer are determined by their respective land cover, soil property
maps for openLISEM should be rather based on the land cover map than on the slope position map.
Secondly, it can be expected that intensive bioturbation just takes place in the most upper soil layer and
therefore that SHPs of soil layers below are mainly determined by texture. Hence, the introduction of a
second soil layer with SHPs predicted with PTFs seems an obvious choice.
4.2. Land cover analysis
In this subsection, the results of the land cover classification and land cover change analysis are presented.
The land cover of Ban Da Na Kham watershed was mapped for the years 2005 and 2018 and subsequently
the appeared land cover changes identified. In the course of this section, the following research questions
are addressed:
• Which land cover changes occurred in the study area between 2005 and 2018?
• What are the possible reasons behind these land cover changes?
The mapping activities were done in GEE employing RF as a classifier and using Landsat 5 and 8 imagery
(Section 3.3). Main land cover classes identified during the fieldwork, which were also used for the
classification, included orchard, cropland, teak plantation, mixed forest and built-up areas.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
43
Orchards are represented by a mix of different perennial crop trees such as long kong, durian and banana.
Long kong and banana are the predominant trees within that class and their plantation area extents from
the valleys up to the slopes until the edges of the forest. It appears that a continuous expansion of the
orchard plantations is taking place. Young fruit trees are planted between the natural trees in an agroforestry
system leading to fuzzy boundaries of the two classes. At the same time, natural trees were noticed being
carved to cut off the nutrient supply in order to ultimately let the trees die. Cropland is mostly situated in
the wider valleys in the low-lying land, where the main streams are located thus simplifying irrigation.
Flooded rice paddy fields hold the majority share in this class. Apart from that, corn and beans were
cultivated. Interesting is the distribution of the teak plantations, being arranged in higher altitudes
surrounding the watershed, areas characterized by clay-rich soils (Section 4.1.3). This might be related to the
fact that soils with high clay content support the growth of teak trees as discussed by Pramono et al. (2015).
Teak as a tropical hardwood tree species is grown for high-quality timber production and represents one of
the livelihoods of the local population. The trees are planted in a line pattern to ease plantation maintenance
(Pramono et al., 2015). Urban areas can be found in two different patterns and are located in the main valleys
of the research area (Figure 12). Firstly, as isolated clusters, each featuring one village community, and
second as buildings lining the roads mostly coupled with shops where local products such as harvested fruits
and wooden goods are sold. In Table 23 the land cover area distribution among the different classes for
both years are shown.
Table 23. Land cover area for 2005 and 2018.
Land cover Area 2005 (km2) Area 2018 (km2)
Orchard 34.71 39.80
Cropland 2.02 3.12
Teak Plantation 4.73 5.46
Mixed Forest 45.13 36.91
Urban 0.32 1.60
The accuracy assessment of the land cover map for 2018 yielded an overall accuracy of 89.2 % and a Kappa
coefficient of 0.86 (Table 24). According to Rwanga et al. (2017), a Kappa of >0.81 signalizes a very strong
agreement. The weakest producer accuracy of 66.9 % was found in the cropland class, where 6 out of 40
reference points were classified as orchards (Table 24). One reason for this misclassification could be an
early stage orchard, represented by young trees with a not yet fully developed leaf canopy. In this case, most
of the reflection of the pixel would originate from the soil surface rather than from the fruit tree itself.
Another possible explanation could be a freshly cut banana plantation, leading equally to extensive soil
background noise. This could easily be associated with a cropland area, which can be for example, brownish
coloured when the paddy is ripe. Similar misclassifications were observed in the orchard class, with 8
reference points classified as cropland (Table 24).
Fuzzy boundaries between orchards, mixed forests and teak plantations are expected to be the main reason
for wrongly classified pixels in the mixed forest and teak plantation class. Reference pixels of the teak
plantation class, classified as cropland can be linked to faded teak blossom and surface cover of brownish
coloured foliage, occurring during the dry season. Looking at the urban class, confusions appeared with
orchards and cropland pixels. This might be attributable due to the close vicinity of the urban area to the
orchards and croplands, thus potentially causing propagation of mixed pixels. Furthermore, similarities
between the colours of rooftop materials and cropland areas are expected to partly cause such confusions.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
44
For 2005 a larger area was classified in order to simultaneously create a land cover map for (I) Ban Da Na
Kham watershed (Figure 12) and (II) Laplae watershed (Annex 8). The accuracy of the land cover maps for
2005 is 87.8 %, with a lower but still strong Kappa coefficient of 0.84 compared to 2018 (Table 25). Lowest
producer accuracy (77 %) was obtained in the orchard class where confusions mainly happened with pixels
of the mixed forest class (Table 25). This might be due to the less developed orchard plantations compared
to 2018. Data for 2005 indicate that fruit trees were mostly still situated in the bottom part of the narrow
tributary valleys, where they were easily covered by the leaf canopy of the mixed forest. In 2005, a water
class had to be added into the classification procedure as a small lake, and several larger ponds were identified
in Laplae watershed (Annex 8). In the water class confusions mostly appeared with urban areas and cropland.
The former might have two reasons, firstly the proximity of roads and buildings to the lake and secondly
blue coloured corrugated iron roofs from some of the buildings. Water, classified as cropland, might be
caused by small ponds located on the agricultural areas or by flooded paddy fields where the rice plants not
yet represent the majority of the observable surface cover. Remaining misclassifications are similar to the
ones from the land cover map of 2018 and were discussed above.
Attempts for further improvement of the classifications could involve different actions. Various studies
focusing on land cover mapping emphasized the exploitation of auxiliary information for classification
accuracy improvements (Saah et al., 2019; Khatami et al., 2016; Zhu et al., 2016; Sluiter & Pebesma, 2010;
Franklin & Wulder, 2002). In this study, only ancillary information extracted from elevation datasets and
optical imagery were used (Section 3.3.4). Additional information for example distances to roads and
buildings obtained from sources such as OSM (Saah et al., 2019), or the incorporation of texture layers, are
additional popular ways to yield classification improvements (Coburn & Roberts, 2004). In order to forestall
confusions between classes especially between classes characterised by trees, such as mixed forest, teak
plantation and orchards, the inclusion of canopy cover and tree height as covariate layer could have also
been supportive as shown by Saah et al. (2019). Canopy cover information could be retrieved from Landsat
products using a regression and classification tree procedure (Hansen et al., 2011). A method for tree height
estimation is presented by Hansen et al. (2016), who used besides of Landsat time-series multi-spectral data,
GLAS (Geoscience Laser Altimeter System) also height data as input for their regression tree model.
Figure 12. Land cover map Ban Da Na Kham watershed, a) 2005 and b) 2018.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
45
Table 25. Accuracy assessment of the land cover map 2005.
Classified Image Reference Data
Row
Total
User Accuracy (%) Cropland Mixed Forest Orchard Teak
Plantation
Urban Water
Cropland 86 5 2 93 92.5
Mixed Forest 1 91 22 5 1 1 121 75.2
Orchard 5 3 77 4 1 90 85.5
Teak Plantation 2 1 35 38 92.1
Urban 1 87 2 90 96.6
Water 29 29 100
Column Total 92 97 100 40 97 35 461 Total Accuracy 87.8 %
Producer Accuracy (%) 93.5 93.8 77 87.5 89.7 82.9 Kappa coefficient 0.84
Additionally, it would have been interesting to investigate the influence of the independent variables on the
prediction of each of the land cover classes in order to assess the relative importance independently.
However, this would require the application of the random forest model to each class (class versus other),
which could not be accomplished in this research study due to time constraints.
4.2.1. Land cover change analysis
Post-classification change detection comparison revealed that substantial changes in different land covers
took place in Ban Da Na Kham watershed over the study period (Figure 13). Changes included urban
developments, deforestation activities and expansion of cultivated areas. The relative changes of each land
cover class within the period between 2005 and 2018 are depicted in Table 26. The greatest increase took
place in the urban class with 394.31 % (1.28 km2), followed by cropland with 54.55 % (1.10 km2) and teak
plantation with 15.55 % (0.74 km2), as well as in the orchard class with 14.68 % corresponding to 5.10 km2.
Mixed forest was the only category where a net decline occurred with 18.2 %.
As mentioned above the largest change in terms of area procent was found in the urban class (Table 26).
This is partially due to urban development in terms of an increase in buildings on the expenses of orchards
(0.77 km2), mixed forest (0.38 km2) and cropland (0.16 km2). Nevertheless, the greatest change can be
attributed to the expansion of the main connecting road between Uttaradit province and the neighbouring
province Phrae, crossing the research area from north to south. The road was extended from originally two
to four lanes (Figure 14). Therefore, the constructed highway now contributes to the urban class in the
classification since its width exceeds 30 m. Thus, is it recognisable on the Landsat images having a spatial
resolution of 30 m. For the classification of 2005, it was not possible to map the existing smaller road with
Table 24. Accuracy assessment of land cover map 2018.
Classified Image Reference Data Row
Total User Accuracy (%)
Cropland Mixed Forest Orchard Teak Plantation Urban
Cropland 33 8 2 2 45 73.3
Mixed Forest 46 2 3 51 90.2
Orchard 6 2 60 2 4 74 81.1
Teak Plantation 1 59 60 98.3
Urban 1 1 83 85 97.6
Column Total 40 49 71 66 89 315 Total Accuracy 89.2 %
Producer Accuracy (%) 66.9 93.9 84.5 89.4 93.3 Kappa coefficient 0.86
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
46
this pixel size as its spectral reflectance got lost due to the reflectance of other land covers represented in
the same pixel. Additionally, the road was often covered by the leaf canopy of trees lining the roadside,
which led to mixed pixels (Figure 14). Hence, the recorded change in Table 26 was, in reality, much smaller
than measured based on the land cover maps.
Changes in cropland seem to have different reasons. As indicated in Table 26, the increase of cropland area
is mainly due to the conversion of orchards (1.22 km2) and mixed forest (0.2 km2). Thereby, the change
from orchards to cropland might be attributed to a rotational cropping system. As it is very unlikely that
fruit trees such as long kong were cut down, it can be expected that those areas were previously occupied
by banana plantations which got pared back. Furthermore, as outlined in section 3.3.9, the results of the
change analysis depend on the classification accuracy. Hence, an exacerbated increase in the area could also
be caused by class confusions, which indeed happened between the cropland and orchard class as described
above.
When comparing the land cover maps of the Ban Da Na Kham watershed of the years 2005 and 2018 a
rivalry of the mixed forest and orchard class in terms of area coverage can be noticed (Figure 12). In 2005,
mixed forest occupies most of the watershed with 51.9 %, whereas in 2018 orchards overtook mixed forest
for being the most prominent land cover with 45.8 % (Table 26). Where the mixed forest partly spread out
towards the valley bottoms in 2005, in 2018 the class is mostly forced back to the steeply sloping terrain and
the summits while experiencing as only class a total area reduction of 8.21 km2 (Figure 12). The land cover
a)
b) Figure 14. Main road in 2005 a) and 2018 b); Source Google Earth Pro.
Figure 13. Change map of Ban Da Na Kham watershed.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
47
change analysis reveals that the large decrease of mixed forest was caused by deforestation followed by
conversion to orchards (15.48 km2) and teak plantations (2.27 km2) (Table 26). Similar observations were
reported by Baicha (2016) who investigated land use dynamics in the Nan province which borders on the
north-east of the Uttaradit province, which has a comparable terrain as the Ban Da Na Kham watershed.
In the Nan province, the natural forest decreased by 3558.07 km2 between 1995 until 2012 having the
greatest reduction in the period between 2009 and 2012 with 2550.08 km2, which corresponds partly to the
period examined in this research. In the same period perennial crops and orchards experienced their largest
expansion with together more than 1600 km2 (Baicha, 2016). Those findings and the findings of the present
study contradict with the Forestry Act from 1989 by the Government of Thailand enforcing a ban on
commercial deforestation at the expenses of natural forest (FAO, 2009). Especially since the Forestry Act
states that all areas with slopes of 35 % and above need to be preserved as forest in order to prevent surface
runoff generation and erosion (FAO, 2009). Nevertheless, during the fieldwork for this study occupation of
steeply sloping terrain by teak and fruit trees were observed.
Increasing by an area of 5.09 km2 orchard was the land cover category increasing the most (Table 26). While
new orchards were created, previous orchards became abandoned (Table 26). Similarly, areas previously
cultivated with fruit trees got reconquered by nature. Thus, new mixed forests emerged (Table 26). The
rationale behind such an interchange could be a hydrometeorological disaster like the one in 2006, where
vast amounts of landslides destructed the landscape, making areas inaccessible for farming (Boonyanuphap,
2013; Usamah et al., 2013).
Two potential reasons explaining the increase in teak plantations were identified. Firstly, the conversion
from orchards (0.21 km2) and cropland (0.01 km2) might be related to environmental policies which were
enforced to reduce agricultural pressure on upland soils. According to Forsyth (2007) past reforestation
activities (teak and pine) conducted by both governmental authorities and non-governmental organizations
were encouraged with the aim to counteract runoff generation and erosional processes by protecting the
soil surface. Furthermore, these plantations were supposed to decrease the need for farming activities as
new livelihood opportunities in the form of plantation maintenance were created (Forsyth, 2007). Secondly,
the increase of teak plantation on the expenses of natural forest (2.27 km2) might be attributable to illegal
logging activities and subsequent reforestation.
Table 26 Land cover change matrix of Ban Da Na Kham watershed (2005 to 2018).
Year Land Cover 2018 (km2) Total (%)
Orchard Cropland Teak Plantation Mixed Forest Urban
2005 (
km
2)
Orchard 23.72 1.22 0.21 8.78 0.77 34.71 39.9
Cropland 0.24 1.61 0.01 0.00 0.16 2.02 2.3
Teak Plantation 0.33 0.07 2.98 1.33 0.02 4.73 5.4
Mixed Forest 15.48 0.20 2.27 26.79 0.38 45.13 51.9
Urban 0.03 0.02 0.00 0.00 0.28 0.32 0.4
Total 39.80 3.12 5.46 36.91 1.60 86.91
(%) 45.8 3.6 6.3 42.5 1.8
Change (km2) 5.09 1.10 0.74 -8.21 1.28
Change (%) 14.68 54.55 15.55 -18.20 394.31
For conclusion the results of the land cover change analysis confirm the trend of converting natural habitats
into agricultural land and urban areas which was also reported by other studies in different region of the
world e.g. Baicha (2016), Barasa and Perera (2018), Lin and Wei (2008), Panahi et al. (2010) and Sajikumar
et al. (2015).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
48
4.3. Flash flood modelling
The focus of this subsection is to address the questions of how different soil information influence the flow
dynamics of the flash flood event of the 22nd of May 2006. Therefore, openLISEM was run with (I) detailed
FD, and (II) SG data. In the second part of this section, the influence of land cover changes in Ban Da Na
Kham watershed on flash flood dynamics is investigated. Prior, the generated input data is briefly described.
In this section, the following research questions will be addressed:
Objective 1: Comparative analysis of SG versus FD for flash flood modelling;
• Is model calibration based on historical flood marks from a nearby watershed possible?
• What are the quantitative differences of the model output using, (I) FD, and (II) SG in relation to
flow dynamics?
• What is the sensitivity of the flood dynamics to different soil depths using, (I) FD, and (II) SG?
Objective 2: Analysis of the effects of land cover change on flash flood behaviour;
• Which land cover generates the highest average runoff?
• How do these land cover changes affect the flood dynamics?
4.3.1. Final model parametrisation
The input data to run openLISEM were generated in the data preparation phase (Section 3.4.1). In the
following, the results of the rainfall analysis and the final soil layer parametrisation are outlined.
Rainfall
Based on Gumble analysis, an RP of 69 years was determined for the event on the 22nd of May 2006 (Annex
9). Since no reliable information about the exact duration of the rainstorm was available, the design storm
was created based on the maximum rainfall depth of 263.7 mm which was recorded on the day of the event.
This resulted in a storm lasting 250 minutes, having a peak intensity of 236 mm h-1 (Figure 15). These
intensities seem to be high, but no reference intensities recorded in this time scale could be found for
validation. However, ADPC (2006) connected the event to a low-pressure system which had developed due
to the cyclone Chan Chu, which devastated the Dan Nag province in Central Vietnam simultaneously. This
process could be a possible explanation for such an extreme rainfall event.
0
50
100
150
200
250
10 30 50 70 90 110 130 150 170 190 210 230 250
Inte
nsi
ty (
mm
)
Duration (min)
Figure 15. Design storm rainfall graph.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
49
Soil data
The results of the soil analysis in Section 4.1 suggested that soil physical properties of the upper layer are
rather influenced by the land cover than by slope position. Hence, the land cover map was used as a mapping
base to generate soil property maps for openLISEM. The final parametrisation of SL1 for the FD setup can
be found in Table 27. Analysing this data shows that mixed forest and teak plantations possess the highest
rates of Ks (>500 mm h-1) and cropland and urban areas the lowest values (<1 mm h-1). Ks in urban areas
was parametrized based on the assumption that those areas are characterised by surface sealing and
compaction. This choice is supported by studies concerning urban hydrology by, for example, Gregory et
al. (2006) and Ossola et al. (2015). Urban porosity values were generated based on the average porosity of
the other classes. Results of SL2 are shown in Table 28 with a gradual increase in Ks going from the summits
(5.6 mm h-1) to the valleys (10.19 mm h-1). The slightly higher porosity of 0.4 cm3 cm-3 on summits can be
linked to the average texture class being clay loam (Table 28). The Psi for the texture classes clay loam
(Summit) and loam (Slope and valley) were calculated with 28.1 cm and 22.27 cm, respectively. For SG
continuous soil property layers were generated. For SG SL1 Ks and porosity range from 81-15 mm h-1 and
0.68-0.51 cm3 cm-3, respectively. Whereas in SG SL2, Ks ranges from 13.4 - 0.15 mm h-1 and porosity from
0.54 - 0.39 cm3 cm-3.
Soil depth:
Using the approach of Kuriakose et al. (2009), four soil depth maps were generated based on topographic
factors. The depth scenarios are compiled as follows: D1 (min: 0.36 m; max: 2.04 m), D2 (min: 1.36 m; max:
3.04 m), D3 (min: 2.36 m; max: 4.04 m), D4 (min: 3.36 m; max: 5.04 m). In Figure 16, an example soil map
(D1) is shown. Larger soil depths are given in the major valleys and on the foot of large hillslopes. Shallow
soil can be found on steeply sloping terrain predominantly in the north-eastern corner of the study area.
Table 27. Soil hydraulic properties SL1.
Land Cover Ks (mm/h) Porosity (cm3/cm3)
Orchard 413 0.54
Cropland 0.208* 0.42
Teak Plantation 593.7 0.51
Mixed Forest 722 0.61
Urban 0.4 0.52
* Wijaya et al. (2009), Ks = Saturated Hydraulic Conductivity
Table 28. Soil hydraulic properties SL2 and
texture class per slope position.
Slope Position Ks (mm/h) Porosity (cm3/cm3) Texture Class
Summit 5.6 0.4 Clay Loam
Backslope 7.27 0.39 Loam
Valley 10.19 0.39 Loam
Ks = Saturated Hydraulic Conductivity
Figure 16. Soil depth map for scenario D1.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
50
4.3.2. Calibration and the effects of different elevation models
Analyzing the research data revealed that a calibration based on historical flood height measurements was
not achievable for various reasons. With none of the employed DEMs, it was possible to generate a flood
pattern comparable to the measured flood heights in Laplae. All the DEMs utilized produced rather sobering
results with unrealistic flood heights of more than 10 m up to 23 m in the mountainous areas (Figure 17).
Simulations with DEMv yielded the highest agreement in terms of points flooded with 44 %. DEMm
achieved 40 % and DEMo 36 % (Annex 10). Simulations with the SRTM DEM had the lowest agreement
with 24 % (Annex 10). However, all simulations showed consistently, that wide parts of the Laplae city
remained dry, and areas which were previously flooded in 2006 received only a few centimetres of water. In
other locations, an overestimation of several meters was simulated. Those mismatches are expected to be
caused by the poor penetration of the C-band radar which applies besides of vegetation also for structures
such as buildings. Hence, in the DEM, urban areas will exhibit elevations related to the top of the roofs
rather than to the road surfaces. Therefore, flood waters must first overcome the approximate height of the
buildings before other areas within the city can be inundated.
In general, the flood distribution pattern of DEMo and DEMm are alike. Using DEMm drainage of flood
waters is slightly better compared to DEMo. However, both simulations still show large accumulations of
flood waters in the upper mountainous valleys (Figure 17). In agricultural areas in the southern lowland, the
flood spreads in a lake-like pattern having water depths of up to 10 m (Figure 17).
Using DEMv and the SRTM DEM the flood distribution is controlled by small local depressions. However,
simulations with the SRTM DEM indicate a much greater dispersion of the flood water, especially in the
lowland (Figure 17). In the simulation with DEMv northern mountainous areas show a flood pattern similar
to DEMo and DEMm but with reduced intensity. In the lowland areas, floods are comparably small and with
shallow water depths (0-1 m) (Figure 17). This might be due to improved drainage in the cropland area
caused by the vegetation corrections.
Rice fields possess, in general, a flat ground. However, the modelling revealed various water accumulations
within the cropland area leading to large ponds with water depths of several meters (Figure 18). It is assumed
that isolated orchards, natural trees and bushes surrounding the paddy fields might affect the radar and
therewith causing errors in the DEM. Hence, it can be concluded that this represents potentially the main
reasons for these artificial ponds.
Another reason which possibly impeded the calibration attempts is the complexity of the 2006 flash flood
event. Apart from the flooding, also debris flows and landslides occurred in the research areas
(Boonyanuphap, 2013). Both are processes which can have a substantial effect on flood dynamics (Marchi,
2017). Therefore, openLISEM might not be the best choice to model such a complex situation. A solution
could represent a multi-hazard model which is able to model hazard interactions. However, the drawback
of such a model is the amount of detailed data required (Bout et al., 2018).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
51
Overall, it appears that problems caused by incorrectly detected elevations caused by the top of vegetation
and rooftops create a DEM which shows many small depressions that are likely not present in reality, and
that have a great influence on the flood behaviour. These potential errors could not be solved by the
corrections applied. This effect not only affects the local flood depth, but also the flow connectivity: water
that is stored in upstream locations will not continue to flood the area that was flooded in reality. Hence, a
realistic representation of the flood situation in 2006 was not possible with any of the DEMs. All DEMs
result in unrealistic flood depths, in wrong locations and too shallow or too deep (more than 10 m). DEMo
and DEMm simulate the most realistic flood pattern in terms of spread. In contrast, SRTM DEM performs
worst.
Figure 17. Flood simulation results in Laplae watershed using different DEMs: a) DEMo; b) DEMv; c) DEMm and d) SRTM DEM.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
52
4.3.3. Sensitivity of flood dynamics to soil information
Eight model simulations were run to assess the sensitivity of flood dynamics in terms of (I) soil data source,
and (II) soil depth. Additionally, the differences between SG and FD were calculated for each soil depth
scenario. Employing SG as input much higher flood volumes were simulated for all three soil depths with
28.0 (D1), 27.2 (D2), 26.5 (D3) and 25.5 (D4) versus 24.0 (D1), 19.5 (D2), 14.2 (D3) and 9.0 million m3 (D4)
(Table 29). This is caused by the infiltration behaviour, reflected by the amount of infiltration and overall
runoff percentage generated. Runoff percentage was calculated as total outflow/total precipitation. Utilizing
SG data, greater flood extents are simulated in comparison to FD with an increase of 8 %, 25 %, 55 % and
110 % for D1, D2, D3 and D4, respectively (Table 29). Using SG, the overall peak discharge is considerably
higher for each soil depth scenario and the differences become more significant with each meter of
additional soil (Table 29). However, in general, peak discharges are decreasing for both soil data sources
with increasing soil depth. Using FD, the decrease is considerable larger with -62 % (D4 – D1) compared
to SG with -27 % (D4 – D1) (Table 29). Similar behaviour is observable for the runoff percentage. In this
case, the lowest runoff percentage simulated for SG in scenario D4 is still 6 % larger than the highest runoff
percentage in FD-D1 with 48 % (Table 29).
Reasons for these observations are different mechanisms set into function for each soil data source. When
using FD, the hydrological behaviour of the watershed is controlled by the storage capacity of the soil,
including porosity and soil depth due to high infiltration rates of the porous SL1. Runoff is generated when
SL1 reaches saturation, and rainfall intensities exceed the infiltration rate of SL2. In contrast, using SG, the
hydrology of the watershed is predominantly determined by the infiltration rate and only to a limited extent
by the soil storage. Correspondingly, during the rainfall, the rain intensity will already exceed the lower
infiltration rates of SL1 and runoff is generated, even though the soil might still have empty pores.
Figure 18. Example of DEM irregularities in cropland areas; a) section of land cover map, b) flood simulation with deep water ponds in parts of the cropland, c) hillshade showing undulating surface of cropland area, d) photo taken with view in north-east direction standing at black dot c).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
53
Table 29. Water balance for eight model simulations using field data (FD) and SoilGrids (SG) data by using four soil
depth scenarios; (SG-FD) represents the difference between SG and FD for each soil depth scenario.
Watershed Parameters FD SG D1
(SG-FD)
D2
(SG-FD)
D3
(SG-FD)
D4
(SG-FD) D1 D2 D3 D4 D1 D2 D3 D4
Total Infiltration (mm) 62 102 142 183 15 25 35 45 -76% -75% -75% -75%
Total outflow (mm) 127 90 61 38 172 162 153 143 35% 80% 151% 276%
Runoff (%) 48 34 23 14 65 61 58 54 35% 80% 151% 276%
Flood vol. (million (m3))* 24 19.5 14.2 9 28 27.2 26.5 25.5 17% 39% 87% 183%
Flood area (km2)* 11.8 10.1 8 5.8 12.8 12.6 12.4 12.2 8% 25% 55% 110%
Peak discharge (thousand
(l/s)) 214 154 104 81 343 312 264 249 60% 103% 154% 207%
*Values correspond to a flood threshold (flow height reported as flood) of 0.3 m, D1 = min: 0.36 m; max: 2.04 m, D2= min: 1.36 m; max: 3.04 m, D3= min: 2.36 m; max: 4.04 m, D4= min: 3.36 m; max: 5.04 m
The simulated flood distribution pattern when using FD and SG is similar and follows the terrain. Flooding
primarily takes place along main channels and in the wide valleys (Figure 19). In lowland area in the south
where cropland is the predominant land cover, flood depths range from 0.5 m – 5 m, whereas 1 – 3 m is
the class occupying the largest area (Figure 19). Those comparable low flood depths can be linked to the
wide geometry of the terrain enabling flood water spread, and to wider channels which ensure more
discharge. In the northern part of the watershed in higher altitudes flood depths are increasing having mostly
more than 3 m and often more than 10 m, due to reduced drainage in narrow valleys. In the tributaries,
flood waters are shallower (0.3-1 m) as either infiltration or drainage to the main channels takes place (Figure
19). Apart from that, isolated very deep lakes with water depths of more than 10 m can be observed (Figure
19). Those water accumulations are formed by a lack of drainage connectivity, introduced by the poor quality
of the DEM as discussed in section 4.3.2. Nevertheless, occasional natural sinks possibly capable of storing
a few meters of storm water were observed during the fieldwork but not in the extent simulated.
Although the flood patterns are similar, simulations with SG lead to recognisable larger and deeper floods
throughout the watershed (Figure 19). Table 30 presents the changes in area related to the flood depth
classes with increasing soil depth for FD and SG. Obviously, the flooded area is shrinking with deeper soil
independently from the soil data source (Table 30). Interesting, however, is which flood depth classes are
affected most. For instance, using SG, areas with comparable shallow floods (0.3-1 m) experience the least
total decrease with 0.5 % whereas in case of FD the same class is reduced by 45.3% (Table 30). In scenario
FD-D4, the flooded area is very limited and mostly represented by shallow flood levels where formerly areas
with large water depths could be found (Table 30). In contrast, simulations with SG show a minimal decrease
at all flood levels. In scenario D4, there are 1.16 km2 still occupied by water heights between 5 and 10 m and
even 0.15 km2 with >10 m (Table 30). This underlines the great sensitivity of the FD in regards to soil depth
information, due to its ability to decide between a large-scale watershed-wide flood (D1) and small-scale
localised floods when using D4 (Figure 19). SG continuously overestimates flood extent and depth. The
effects of soil depth seem to be minor. Nevertheless, the sensitivity of SG to soil depth information should
not be underestimated even it appears weak in comparison.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
54
Figure 19. Maximum flood depth (m) for field data (FD) and SoilGrids (SG) for different soil depths: a) FD-D1, b) FD-D2, c) FD-D3, d) FD-D4, e) SG-D1, f) SG-D2, g) SG-D3 and h) SG-D4, D1= min: 0.36 m; max: 2.04 m, D2= min: 1.36 m; max: 3.04 m, D3= min: 2.36 m; max: 4.04 m, D4= min: 3.36 m; max: 5.04 m.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
55
Flood duration is influenced by the terrain. In areas, with high flood waters, the duration exceeds 10 h and
often stays more than 60 h (Figure 20). Therefore, simulations with SG show prolonged flood durations
compared to FD, caused by less and slower infiltration, more runoff and thus more water which needs to
evacuate.
The dependency of flood duration related to the terrain is especially visible in mountainous pits where flood
waters got locked and remained for more than 60 h (Figure 20). In wide cultivated valleys with flood depths
between 0.5 – 5 m floods last long as well (Figure 19 & Figure 20). The low flow velocity is due to the almost
flat terrain and therefore, slow drainage (Figure 20). This behaviour is intensified by openLISEM as channel
inflow is controlled by surface flow velocity, which is often almost 0 m s-1 in those areas. It appeared that
the channels did not always transport water with their full capacity even if they passed deeply flooded areas.
Table 30. Flood depth related to soil data information and total reduction from D1 to D4.
Flood depth (m) FD (km2)
Total
(D4-D1)
SG (km2) Total
(D4-D1) D1 D2 D3 D4 D1 D2 D3 D4
0.3 5.4 4.8 3.9 3.0 -45.3% 5.52 5.53 5.51 5.49 -0.5%
1 3.8 3.2 2.6 2.0 -47.4% 4.2 4.1 4.0 3.9 -6.4%
3 1.4 1.2 0.95 0.6 -56.2% 1.7 1.61 1.55 1.52 -8.6%
5 1.1 0.8 0.5 0.2 -78.6% 1.3 1.19 1.21 1.16 -11.2%
>10 0.1 0.09 0.04 0.01 -96.0% 0.2 0.17 0.16 0.15 -14.0%
FD= Field Data, SG = SoilGrids D1= min: 0.36 m; max: 2.04 m, D2= min: 1.36 m; max: 3.04 m, D3= min: 2.36 m; max: 4.04 m, D4= min: 3.36 m; max: 5.04 m
Figure 20. Maximum flood duration (hr) using field data a) and SoilGrids b) with soil depth scenario D1 (min: 0.36 m; max: 2.04 m). Zoomed area shows how flood time changes when using different soil information.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
56
In contrast, short flood durations can be found in tributaries, in areas with steeply sloping terrain and shallow
flood levels as both conditions promote quick drainage (Figure 19 & Figure 20). Nevertheless, also areas in
the northern parts of the watershed, indicate long flood durations even in parts characterized by steep slopes
which should counteract flood accumulations. These simulated hydrological behaviours can be attributed
to the problematic DEM as surface water is erroneously slowed down and finally blocked in the middle of
the slope.
Infiltration is increasing with each additional meter of soil added (Table 29). This is the case for both FD
and SG. However, it is worth to note with each meter of soil added, the average infiltration of the FD is
increased by approximately 40 mm whereas the average infiltration utilizing SG is always approximately 75%
less for each soil depth scenario (Table 29).
Figure 21 shows infiltration maps for both soil datasets under D1 scenario. Using FD, it is observable that
the infiltration pattern depends on the soil depth distribution. After a careful investigation, it appears that
areas with high infiltration values (70 – 85 mm) are predominantly found in wider valleys and occasionally
on summits (Figure 21). The former shows higher infiltration values for two reasons, (I) water from upslope
areas is accumulated, and (II) soil is deeper. This leads inevitably to more infiltration capacity. However, in
valleys also variations in infiltration can be observed, depending on their location in the terrain and the
predominant land cover. In the main valleys where urban areas are situated, and crops are cultivated, the
infiltration is low (0-40 mm) due to compaction of SL1. Correspondingly the available storage cannot be
fully exploited (Figure 21).
Figure 21. Total infiltration (mm) for field data a) and SoilGrids b) using soil depth scenario D1 = (min: 0.36 m; max: 2.04 m). Zoomed section of a) shows high infiltration values next to impermeable areas. Zoomed section of b) shows artificial border between low and high infiltration values.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
57
Summits experience less infiltration since water is in this case only received by rainfall and not by runoff.
Additionally soils are shallower. Variations of infiltration on summits vary between 60-85 mm caused by the
difference in soil depth and to a small extent by changes in land cover (Figure 21). The rather low Ks of SL2
(5.6 mm h-1) on summits seems to be counteracted by high flow resistance values of the teak plantations
and mixed forest, thus reducing the flow velocity of surface water and promoting infiltration.
On slopes, more runoff is generated, and infiltration is decreased. This effect is caused by shallow soil
depths, which limit infiltration and an increasing slope gradient, which accelerates runoff flow velocities.
Additionally, the predominant cultivation in terms of fruit trees in sloping areas causes less infiltration due
to its low flow resistance values. Consequently, besides of the limited infiltration capacity, the surface flow
velocity is enhanced.
Nevertheless, almost all areas possess infiltration values > 40 mm (Figure 21). The porous upper soil layer
with high Ks enables absorption of even high rainfall intensities, only limited by the available pore space.
Large amounts of runoff generated on compacted surfaces (built-up and cropland) evacuate to adjacent
pixels causing very high infiltration values of more than 85 and even up to 505 mm (Figure 21). On the one
hand, this behaviour is in line with the hydrological principles; on the other hand, those values are too high
and therefore might be caused by numerical errors within openLISEM.
Infiltration of SG is considerably lower compared to the results of the FD (Figure 21). The infiltration
pattern is clearly determined by the distribution of SGs SOM and Db in the research area (Annex 11).
Furthermore, high infiltrations (>20 mm) can be found predominantly in the north-eastern part of the
watershed, in the higher altitudes at the foot of the great northern slopes, defining the upper watershed
border (Figure 21). Those areas are spots of runoff accumulation being characterised by low slope gradient,
which favours infiltration. Interestingly a small spatial variability of SG is visible. This is shown as areas with
low infiltrations (8 -13 mm) correspond very roughly the distribution of built-up areas and cropland (Figure
21). Both are characterised by high Db and therefore, lower Ks and reduced porosity (Annex 11).
Analysing the “sub-patterns” reveals, that these are comparable to the ones produced by the FD. Thus, it
can be argued that SGs infiltration seems additionally influenced by slope gradient and associated soil depth.
Therefore, it is expected that the same mechanisms as described above, apply. In addition, in the northern
part of the watershed, several sharp boundaries between areas with high and low infiltration are visible.
Those infiltration patterns do not follow any logic as they do not correspond to any land cover boundary,
nor to the soil depth distribution, and also not to the slope gradient (Figure 21). This pattern seems to be
artificial as such boundaries are not expected to be found in natural landscapes where soil and its properties
change more gradually. The same line pattern can be found on SGs SOM layer and is most likely attributable
to a misprediction (Annex 11).
4.3.4. Effects of land cover change on flood dynamics
Table 31 shows the results of two model simulations for Ban Da Na Kham watershed, which were run in
order to assess the effects of land cover changes on runoff and flood dynamics. In the period between 2005
to 2018, there was an expansion of predominantly orchards on the expenses of mixed forest. Further smaller
changes comprised an increase in urban and cropland areas, as well as an increase in teak plantations (Section
4.2.1). The results contain interesting insights, for example, flood volume and inundated area decreased
from 24.3 million m3 and 12.1 km2 to 24 million m3 and 11.8 km2 from 2005 to 2018 (Table 31).
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
58
Key figures such as total infiltration and runoff percentage experienced neglectable or no changes. Besides,
it appears that the peak discharge increased by 1.9 %, and the peak time of the discharge decreased from
348 to 315 min (Table 31). Thus, those results indicate an accelerated watershed response time and an
increase of storm water transportation due to the conversion from mixed forest to orchards.
Other studies concerning land cover change and its associated effects on watershed hydrology, investigated
similar transitions of forested areas into cultivated and urban land. Regardless of the simulated time scale,
whether they are event-based (Barasa et al., 2018), seasonal (Lin et al., 2008) or perennial (Sajikumar et al.,
2015), they emphasise an increase of peak discharges. Interestingly, a shift in peak time, as observed in this
study was not reported. Sajikumar et al. (2015) detected similar a small increase in peak discharge. They
claimed that similarities of land cover characteristics of classes which underwent the change (forest to
plantation) mainly caused this effect.
In this study, the similarities in the main classes (mixed forest and orchard) are given by their effects on the
underlying soil, reflected by high Ks values in SL1 which ensures full infiltration until saturation state
independent from the class (Section 4.3.1). Differences, on the other hand, are the available pore space in
the soil and the land cover characteristic flow resistance (Sections 3.4.1 & 4.3.1). Quantification of alterations
of SHPs from land cover changes requires actual monitoring of those properties in space and time (Hassler,
2013). Studies from Nyberg et al. (2011), Peng et al. (2012) and Zimmermann et al. (2006), which
investigated the transition of SHPs due to afforestation of cultivated land or vice versa, showed that it can
take several decades until conditions close to full recovery or complete declination are reached, as soil
exhibits some sort of memory effect. However, this process was ignored in this research as differences
between orchards and mixed forest were considered as neglectable small. Furthermore, monitoring of SHPs
is an activity which was technical, not feasible and would have been beyond the scope of this research study.
Runoff can be explained by two theories, infiltration excess and saturation excess. The former assumes
runoff generation as soon as rainfall intensity exceeds the infiltration rate of the soil (Horton, 1933). The
concept of saturation excess, on the other hand, describes runoff initiation caused fully saturated soil (Dunne
& Black, 1970). In this study infiltration rates of SL1 are higher than the rainfall intensities, except for
cropland and urban areas. For SL2 the opposite applies with a maximum Ks of 10.19 mm h-1 in valleys
(Section 4.3.1). Thus, during the modelled rainfall event first SL1 will become saturated as it has limited
storage capacity due to its shallow depth of only 5 cm and almost saturated initial condition. Subsequently,
infiltration and runoff are determined by the infiltration rate of SL2.
Table 31. Water balance of Ban Da Na Kham watershed for 2005 and 2018.
Watershed Parameters Land Cover 2005 Land Cover 2018
Total Infiltration (mm) 63 62
Total outflow (all flows) (mm) 127 127
Runoff (%) 47.7 47.7
Flood volume (in million (m3))* 24.3 24
Flood area (km2)* 12.1 11.8
Peak discharge (in thousands (l/s) 210 214
Peak time discharge (min) 348 315
*Values correspond to a flood threshold (flow height reported as flood) of 0.3 m
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
59
Table 32 shows the runoff amounts per land cover class for both years. The runoff amounts were calculated
as precipitation - (infiltration + interception). It appears that the runoff experienced just a slight increase
(0.6 %) from 2005 to 2018. It is evident that urban areas and cropland exhibit the highest average runoff
amounts (>200 mm), in both years. Furthermore, both classes experience an increase in runoff (Table 32).
In contrast, orchards, teak plantation and mixed forest show a considerably lower average runoff with
approximately 200 mm (Table 32). A decrease in average runoff from 2005 to 2018 is only noticeable in the
orchard and teak plantation class with <1 % for both.
The contribution of cropland area to the total runoff increased by 1.3 %. However, the runoff in cropland
is difficult to quantify as rice fields are flooded during the rice production period, and no runoff as such is
generated in reality. To model the flooded condition of the rice fields, those areas were parametrised as
nearly flat areas, which are almost impermeable with Ks of <1 mm h-1 (Section 4.3.1). Nevertheless, it was
not possible to replicate the storing effect which would become effective due to the bunds surrounding the
fields. In reality, the water which would overflow the bunds due to an exceedance of the storage capacity
could potentially considered as runoff.
Interesting is, that in areas of mixed forest, which is the only category where an area net decline occurred,
the average runoff increased (Section 4.2.1 & Table 32). Associated to the changing land cover, the land
cover distribution over the terrain has also changed. This has implications on the runoff initiation since soil
depth, and with it, the available pore space varies in the terrain. Especially since the initial soil moisture was
assumed to be homogenous, consequently, areas having less pore space and/or shallow soil will generate
more surface water. Therefore, the increase in average runoff in the mixed forest class is based on the
deforestation which predominately took place in the valleys and on the foot of the slopes. Both are areas
which have (I) more storage capacity due to deeper soils, (II) higher Ks values of SL2, and (III) decelerated
surface water velocities. Correspondingly, infiltration is promoted, and less runoff generated. In, 2018,
mixed forest predominantly occurs in higher altitudes, on narrow summits and on steep slopes. Especially
on steep slopes, more runoff is initiated as they possess comparably shallow soils. Additionally, an increasing
slope gradient accelerates flow velocities. Consequently, infiltration is not only limited by the storage capacity
of the soil but also by the enhanced surface flow velocities. This runoff terrain dependency is also
transferable to the other land cover classes explaining changes in average runoff amounts. Despite the fact
that soil moisture is not homogenous in the terrain and therefore the runoff amounts might deviate in reality,
it is expected that the results give a fairly good indication of the land cover runoff relation within the
watershed.
Besides causing alterations in runoff amounts, land cover changes also affected the subsequent flood and
its distribution. This is due to the land cover characteristic flow resistance. Table 33 presents various flood
depth classes with the corresponding inundated area for both years. Mixed forest is characterised by dense
Table 32. Runoff amounts per land cover for 2005 and 2018.
Land Cover 2005 2018
Avg. infiltration (mm)
Avg. runoff (mm)
Runoff (%)* Avg. infiltration (mm)
Avg. runoff (mm)
Runoff (%)*
Orchard 66.3 197.1 74.7 66.5 196.8 74.6
Cropland 28.4 235.1 89.1 26.4 237.1 89.9
Teak Plantation 63.4 199.4 75.6 63.5 199.3 75.6
Mixed Forest 63.1 199.8 75.8 62.3 200.6 76.1
Urban 21.3 242.4 91.9 20.1 243.6 92.4
* Runoff percentage is calculated as average runoff/precipitation
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
60
undergrowth and is parameterised accordingly with a high flow resistance. Associated with the increased
cultivation of orchards having considerably less flow resistance, as their ground is almost continuously
covered by short grass, surface water accumulations in mountainous areas got reduced. Hence, the overall
extent of surface water decreased due to the enhanced drainage. This process is evident as water depths of
<5 m decreased from 21.0 km2 (2005) to 20.4 km2 (2018). In contrast, the water levels increased by ≥5 m
from 11.7 (2005) km2 to 11.9 km2 (2018) (Table 33). This increase is shown in Figure 22 and mainly took
place in the lowland areas.
Table 33. Flood depths and their corresponding extent due to land cover change.
Flood depth (m) 2005 Area (km2) 2018 Area (km2) Change (%)
<0.3 10.1 9.8 -3
0.3 5.52 5.4 -2
1 3.94 3.7 -4
3 1.46 1.43 -2
5 1.04 1.06 1
>10 0.13 0.14 6
Changes in water depth also led to changes in flood durations. While in mountainous areas flood durations
decreased with a reduction in water depth, an increase of the maximum flood durations in the lowland area
is noticeable (Figure 23). Hence, in wide valleys where crop cultivation is practised and most of the urban
settlements can be found flood water often remains for more than 10 h and in various locations also for
more than 60 h (Figure 23). In conclusion, it is inherent that changes in land cover did not affect the flood
dynamics of the watershed considerably, which could be expected as changes from a hydrological point of
view were minor.
Figure 22. Maximum flood depth for 2005 a) and 2018 b). Zoomed in area indicates an increase in maximum flood depth in the lowland area due to land cover changes.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
61
Figure 23. Flood duration for 2005 a) and 2018 b). Zoomed in area indicates prolonged floods in the lowland areas due to land cover changes.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
62
5. CONCLUSION AND RECOMMENDATIONS
This research analysed two soil datasets, (I) detailed field data and (II) SoilGrids to assess their similarity and
to determine how sensitive the flood dynamics are in relation to their soil hydraulic properties when applied
in an integrated flood model, to make a statement if SoilGrids represents a reasonable alternative in data
scarce environments. In the second part, the land cover was mapped for two year, and land cover changes
identified, subsequently the effects of those changes on runoff generation and flash flood behaviour were
investigated.
The results imply that the field data and SoilGrids do not share many commonalities. Collected field data is
highly influenced by land cover, whereas SoilGrids variability is limited throughout the watershed. SoilGrids
is tailored to a global scale and therefore, does not necessarily take local phenomena into account. Hence,
applying SoilGrids should be considered carefully. Especially when used for hydrological modelling on a
watershed scale where differences in water holding capacity or infiltration are decisive. Soil physical
properties such as clay content and bulk density were overestimated by SoilGrids. This, in turn, had great
effects when predicting soil hydraulic properties such as saturated hydraulic conductivity using the
pedotransfer functions of Saxton and Rawls (2006). Having predominantly an underestimation of saturated
hydraulic conductivity led to an increased surface runoff during the modelling as rainfall intensities exceeded
the infiltration rate although infiltration would have taken place in reality. Similarly, the overestimation of
soil organic matter content by SoilGrids led to wrong water holding capacities of the soil. Those
overestimations of soil physical and chemical properties and the subsequent false estimation of soil hydraulic
properties by the pedotransfer functions ultimately had far-reaching consequences on the hydrology. An
alteration in the runoff generation mechanism partly appeared, and flood extent, depth and duration
increased considerably. Increasing soil depth affected both datasets similarly by promoting infiltration and
reducing surface water. However, using field data, the flood dynamics were more sensitive to changes in soil
depth. Overall, it can be concluded that the choice of soil input data (soil hydraulic properties and soil depth)
had a great influence on both the quantity and spatial variability of infiltration which consequently affected
runoff and flood dynamics.
The land cover change analysis confirmed the almost omnipresent trend of converting natural habitats into
cultivated and urbanised areas. Subsequent flood simulations with changed land cover information indicated
a slight intensified hydrological watershed response in the form of increased runoff amounts and greater
peak discharge, as well as a reduction of the peak discharge time. Effects on general flood dynamics were
minor as from a hydrological point of view no essential changes occurred. However, it can be concluded
that national policies such as the Forestry Act enforced in 1989 by the Government of Thailand, aiming to
stop commercial forest exploitation show no measurable effect in the research area. Consequently,
strengthening of local level policy implementation and monitoring is urgently needed to ensure
environmental protection and human security.
Furthermore, this study provides valuable insights into the implications of hydrological modelling in
(tropical) data scarce environments beyond the original objectives of this research study, which might be of
interest for future studies. Firstly, the findings reveal that complications which can arise with freely available
elevation data in densely vegetated areas nearly outweighs the effects of using predicted soil hydraulic
properties, retrieved from soil physical and chemical property layers of SoilGrids. The correct representation
of the terrain is of at least equal importance for realistic flood simulations as soil data. It is not only a matter
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
63
of wrong water depths because of errors in the elevation, but the lack or the alteration in flow connectivity
completely changes the flood dynamics. Without a precise digital elevation model, flood modelling is only
useful in a relative sense (intercomparing of scenarios) but not when used for a flood hazard analysis which
requires an accurate flood location. Secondly, in tropical areas where processes influencing the structure of
soil (e.g., bioturbation) can be expected, it is advisable to develop a soil sampling strategy based on the
prevailing land cover types rather than on the terrain or an existing soil texture map. Since biotic macropores
might have a dominating influence on soil hydraulic properties instead of the soil texture. Thirdly, it is
suggested to further subdivide the land cover class orchard for soil sampling and classification purposes to
prevent soil property aggregation as experienced in this study. Lastly, additional sampling of subsurface soil
(B horizon) is recommended as soil properties are expected to considerably change due to the absence of
surface layer particularities (e.g., bioturbation) and soil hydraulic properties representing only the most upper
layer do not suffice when modelling extreme events, that will have deeper infiltration.
In the course of this research, various limitations had to be faced. Firstly, the ruggedness of the terrain
interfered a well-balanced soil sample distribution, hence summits and steep slopes and with them the mixed
forest and teak plantation class are underrepresented in the soil analysis. Secondly, the representation of
flooded rice fields in the model encountered difficulties. In the study area, the rice fields are surrounded by
bunds having an outlet which controls the water level by enabling drainage during rainfall events. Therefore,
during an extreme event, the rice fields would first function as some sort of reservoir storing the stormwater
before water is released. Those interactions are complicated to model and need detailed investigations in
order to make an adequate representation possible. Hence, in openLISEM, the rice fields had to be
represented in a simplified way, which influenced the flood dynamics in the lowland areas. Lastly, the
greatest limitation of this research is the quality of the DEM, which adversely affected flood dynamics and
thereby led to an obstruction of the calibration efforts undertaken. Correction attempts and smoothing did
not harvest any improvements, resulting in a wrong flood pattern. However, the goal of this research was
not to do a hazard analysis, which would have been a futile exercise with any hydrological model when using
this elevation model. Therefore, as the infiltration process and therefore, the overall runoff quantities can
be expected to be correct, the quality of the digital elevation model does not invalidate the conclusions for
the objectives set.
Future research should continue exploring SoilGrids data applicability for hydrological modelling as it
represents a valuable source of information. To be able to make a profound statement about SoilGrids
quality, it is necessary to conduct studies in various regions of the world. This will help to investigate its
quality dependency on factors such as terrain, climate and vegetation.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
64
List of References
ADPC. (2006). Rapid Assessment: Flashflood and Landslide Disaster in the Provinces of Uttaradit and Sukhothai Northern Thailand. Bangkok. Retrieved from http://www.adpc.net/enewsjuly/Uttaradit_rapidassessment.pdf
Ahuja, L. R., Naney, J. W., Green, R. E., & Nielsen, D. R. (2010). Macroporosity to Characterize Spatial Variability of Hydraulic Conductivity and Effects of Land Management. Soil Science Society of America Journal, 48(4), 699–702. https://doi.org/10.2136/sssaj1940.036159950004000c0132x
Angiuli, E., & Trianni, G. (2014). Urban Mapping in Landsat Images Based on Normalized Difference Spectral Vector. IEEE Geoscience and Remote Sensing Letters, 11(3), 661–665. https://doi.org/10.1109/LGRS.2013.2274327
Aon Benfield. (2012). 2011 Thailand Floods Event Recap Report. Chicago. Retrieved from http://thoughtleadership.aonbenfield.com/Documents/20120314_impact_forecasting_thailand_flood_event_recap.pdf
Arcement, G. J., & Schneider, V. R. (1989). Guide for Selecting Manning’s Roughness Coefficients for Natural Channels and Flood Plains (Water-Supply Paper No. 2339). Retrieved from https://pubs.usgs.gov/wsp/2339/report.pdf
Arrouays, D., Grundy, M. G., Hartemink, A. E., Hempel, J. W., Heuvelink, G. B. M., Hong, S. Y., … Zhang, G.-L. (2014). GlobalSoilMap: Toward a Fine-Resolution Global Grid of Soil Properties. In Advances in Agronomy (Vol. 125, pp. 93–134). Academic Press. https://doi.org/10.1016/B978-0-12-800137-0.00003-0
As-syakur, A. R., Adnyana, I. W. S., Arthana, I. W., & Nuarsa, I. W. (2012). Enhanced Built-Up and Bareness Index (EBBI) for Mapping Built-Up and Bare Land in an Urban Area. Remote Sensing, 4(10), 2957–2970. https://doi.org/10.3390/rs4102957
ASF. (2015). ASF radiometric terrain corrected products. Baicha, W. (2016). Land use dynamics and land cover structure change in Thailand (as exemplified by
mountainous Nan Province. Geography and Natural Resources, 37(1), 87–92. https://doi.org/10.1134/s1875372816010121
Barasa, B. N., & Perera, E. D. P. (2018). Analysis of land use change impacts on flash flood occurrences in the Sosiani River basin Kenya. International Journal of River Basin Management, 16(2), 179–188. https://doi.org/10.1080/15715124.2017.1411922
Bauer, E., & Kohavi, R. (1999). An Empirical Comparison of Voting Classification Algorithms: Bagging, Boosting, and Variants. Machine Learning, 36(1/2), 105–139. https://doi.org/10.1023/A:1007515423169
Bausch, W. C. (1993). Soil background effects on reflectance-based crop coefficients for corn. Remote Sensing of Environment, 46(2), 213–222. https://doi.org/10.1016/0034-4257(93)90096-G
Beers, T. W., Dress, P. E., & Wensel, L. C. (1966). Notes and observations: aspect transformation in site productivity research. Journal of Forestry, 64, 691–692. https://doi.org/10.2307/2653667
Beon, M.-S., Cho, K. H., Kim, H. O., Oh, H.-K., & Jeong, J.-C. (2017). Mapping of Vegetation Using Multi-Temporal Downscaled Satellite Images of a Reclaimed Area in Saemangeum, Republic of Korea. Remote Sensing, 9(272), 2–17. https://doi.org/10.3390/rs9030272
Blanchez, J. (1997). Forest Resources and Roundwood supply in the Asia Pacific Countries : Situation and Outlook to the Year 2010. Asia - Pacific Forestry Sector Outlook Study Working Paper Series. Retrieved from http://nkcs.org.np/dfrs/cfl//opac_css/index.php?lvl=notice_display&id=9114
Bonfante, A., & Bouma, J. (2015). The role of soil series in quantitative land evaluation when expressing effects of climate change and crop breeding on future land use. Geoderma, 259–260, 187–195. https://doi.org/10.1016/J.GEODERMA.2015.06.010
Boonprong, S., Cao, C., Chen, W., & Bao, S. (2018). Random forest variable importance spectral indices scheme for burnt forest recovery monitoring-multilevel RF-VIMP. Remote Sensing, 10(807), 2–15. https://doi.org/10.3390/rs10060807
Boonyanuphap, J. (2013). Cost-benefit analysis of vetiver system-based rehabilitation measures for landslide-damaged mountainous agricultural lands in the lower Northern Thailand. Natural Hazards, 69(1), 599–629. https://doi.org/10.1007/s11069-013-0730-y
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
65
Botula, Y.-D., Van Ranst, E., & Cornelis, W. M. (2014). Pedotransfer functions to predict water retention for soils of the humid tropics: a review. Revista Brasileira de Ciência Do Solo, 38(3), 679–698. https://doi.org/10.1590/S0100-06832014000300001
Boumana, B. A. M., Wopereisb, M. C. S., Kropffc, M. J., Ten Bergea, H. F. M., & Tuongc, T. P. (1994). Water use efficiency of flooded rice fields II. Percolation and seepage losses. Agricultural Water Management, 26, 291–304. Retrieved from https://core.ac.uk/download/pdf/29341530.pdf
Bout, B., & Jetten, V. G. (2017). The validity of flow approximations when simulating catchment- integrated flash floods. Journal of Hydrology, 556, 674–688. https://doi.org/10.1016/j.jhydrol.2017.11.033
Bout, B., & Jetten, V. G. (2018). OpenLISEM Multi-Hazard Land Surface Process Model - Documentation & User Manual. Enschede: ITC Faculty of Geo-Information Science and Earth Observation of the University of Twente.
Bout, B., Lombardo, L., van Westen, C. J., & Jetten, V. G. (2018). Integration of two-phase solid fluid equations in a catchment model for flashfloods, debris flows and shallow slope failures. Environmental Modelling & Software, 105, 1–16. https://doi.org/10.1016/J.ENVSOFT.2018.03.017
Breiman, L. (2001). Random Forests. Machine Learning, 45, 5–32. Retrieved from https://link.springer.com/content/pdf/10.1023%2FA%3A1010933404324.pdf
Bronstert, A., Niehoff, D., & Gerd, B. (2002). Effects of climate and land-use change on storm runoff generation : present knowledge and modelling capabilities. Hydrological Processes, 16(2), 509–529. https://doi.org/10.1002/hyp.326
Brooks, R. H., & Corey, A. T. (1964). Hydraulic Properties of Porous Media. Hydrology Papers, 3, 1–25. Retrieved from https://mountainscholar.org/bitstream/handle/10217/61288/HydrologyPapers_n3.pdf?sequence
Brovkina, O., Zemek, F., Novotný, J., Heřman, M., & Štěpánek, P. (2017). Analysing changes in land cover in relation to environmental factors in the districts of Znojmo and Třebíč (Czech Republic). European Journal of Environmental Sciences, 7(2), 108–118. https://doi.org/10.14712/23361964.2017.9
CFE-DM. (2018). Disaster Management Reference Handbook Thailand. Joint Base Pearl Harbor - Hickam: Center for Excellence in Disaster Management and Humanitarian Assistance. Retrieved from https://www.cfe-dmha.org
Chastain, R., Housman, I., Goldstein, J., Finco, M., & Tenneson, K. (2019). Empirical cross sensor comparison of Sentinel-2A and 2B MSI, Landsat-8 OLI, and Landsat-7 ETM+ top of atmosphere spectral characteristics over the conterminous United States. Remote Sensing of Environment, 221, 274–285. https://doi.org/10.1016/J.RSE.2018.11.012
Chen, J., Chen, J., Liao, A., Cao, X., Chen, L., Chen, X., … Mills, J. (2015). Global land cover mapping at 30 m resolution: A POK-based operational approach. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 7–27. https://doi.org/10.1016/J.ISPRSJPRS.2014.09.002
Chen, L., Van Westen, C. J., Hussin, H., Ciurean, R. L., Turkington, T., Chavarro-rincon, D., & Shrestha, D. P. (2016). Integrating expert opinion with modelling for quantitative multi-hazard risk assessment in the Eastern Italian Alps. Geomorphology, 273, 150–167. https://doi.org/10.1016/j.geomorph.2016.07.041
Chow, V. T. (1959). Open-Channel Hydraulics. (H. E. Davis, Ed.) (Internatio). New York: McGraw Hill Book Company.
Cingolani, A. M., Renison, D., Zak, M. R., & Cabido, M. R. (2004). Mapping vegetation in a heterogeneous mountain rangeland using landsat data: An alternative method to define and classify land-cover units. Remote Sensing of Environment, 92(1), 84–97. https://doi.org/10.1016/j.rse.2004.05.008
Coburn, C. A., & Roberts, A. C. B. (2004). A multiscale texture analysis procedure for improved forest stand classification. International Journal of Remote Sensing, 25(20), 4287–4308. https://doi.org/10.1080/0143116042000192367
Cohen, J. (1960). A Coefficient of Agreement for Nominal Scales. Educational and Psychological Measurement, 20(1), 37–46. Retrieved from https://journals-sagepub-com.ezproxy2.utwente.nl/doi/pdf/10.1177/001316446002000104
Colby, J. D. (1991). Topographic Normalization in Rugged Terrain. Photogrammetric Engineering & Remote Sensing, 57(5), 531–537. Retrieved from https://pdfs.semanticscholar.org/4f0c/45d90bbae073872c2d4cd73aa4fa2e6da4c9.pdf
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
66
Colditz, R. R. (2015). An evaluation of different training sample allocation schemes for discrete and continuous land cover classification using decision tree-based algorithms. Remote Sensing, 7(8), 9655–9681. https://doi.org/10.3390/rs70809655
Comber, A., Balzter, H., Cole, B., Fisher, P., Johnson, S., & Ogutu, B. (2016). Methods to Quantify Regional Differences in Land Cover Change. Remote Sensing, 8(176), 1–19. https://doi.org/10.3390/rs8030176
Congalton, R. G. (1991). A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote Sensing of Environment, 37, 35–46. Retrieved from https://pdfs.semanticscholar.org/d7be/d062683df1fd6723fb9c0c1d26feddc8c133.pdf
Crist, E. P., & Cicone, R. C. (1984). A Physically-Based Transformation of Thematic Mapper Data-The TM Tasseled Cap. IEEE Transactions on Geoscience and Remote Sensing, 22(3), 256–263. Retrieved from http://www.geo.umass.edu/courses/geo594q/Crist1984IEEETasseledCap.pdf
D’Souza, G. (2000). Forest Cover Mapping & Monitoring with NOAA-AVHRR & other Coarse Spatial Resolution Sensors (No. 29). Rome. Retrieved from www.fao.org/fo
De Paz, M., Gobbi, M. E., & Raffaele, E. (2018). Fallen fruits stimulate decomposition of leaf litter of dominant species in NW Patagonia shrublands. Plant and Soil, 425(1–2), 433–440. https://doi.org/10.1007/s11104-018-3590-0
De Roo, A. P. J., Wesseling, C. G., & Ritsema, C. J. (1996). LISEM: a single event physically-based hydrologic and soil erosion model for drainage basins. I: Theory, input and output. Hydrological Processes, 10(April 1995), 1107–1117.
Devia, G. K., & Ganasri, B. P. (2015). A Review on Hydrological Models. Aquatic Procedia, 4, 1001–1007. https://doi.org/10.1016/J.AQPRO.2015.02.126
DHI. (2017). Mike 11 - Reference Manual. DHI. Retrieved from http://manuals.mikepoweredbydhi.help/2017/Water_Resources/Mike_11_ref.pdf
Di Gregorio, A. (2005). Land cover classification system : classification concepts and user manual : LCCS (No. 2). Rome. Retrieved from http://www.fao.org/3/y7220e/y7220e00.htm
Di Gregorio, A. (2016). Classification Concepts Land Cover Classification System Software version 3 (No. 3). Food and Agriculture Organisation of the United Nations. Retrieved from www.fao.org/
Di Gregorio, A., & Jansen, L. J. M. (2000). Land cover classification system (LCCS): classification concepts and user manual. Rome: Food and Agriculture Organization of the United Nations. Retrieved from http://www.fao.org/3/x0596e/x0596e00.htm
Domaç, A., & Süzen, M. L. (2006). Integration of environmental variables with satellite images in regional scale vegetation classification. International Journal of Remote Sensing, 27(7), 1329–1350. https://doi.org/10.1080/01431160500444806
Dunne, T., & Black, R. D. (1970). Partial Area Contributions to Storm Runoff in a Small New England Watershed. Water Resources Research, 6(5), 1296–1311.
Eijkelkamp. (2013). Laboratory-Permeameters Operating Instructions. Giesbeek: Eijkelkamp Agrisearch Equipment.
Faber-Langendoen, D., Tart, D. L., & Crawford, R. H. (2009). Contours of the Revised U.S. National Vegetation Classification Standard. Bulletin of the Ecological Society of America, 90(1), 87–93. https://doi.org/10.1890/0012-9623-90.1.87
FAO. (2009). Thailand forestry outlook study. Bangkok. Retrieved from http://www.fao.org/docrep/014/am617e/am617e00.pdf
Farr, T. G., Rosen, P. A., Caro, E., Crippen, R., Duren, R., Hensley, S., … Alsdorf, D. (2007). The Shuttle Radar Topography Mission. Reviews of Geophysics, 45, 1–33. https://doi.org/10.1029/2005RG000183
Field, A. (2009). Discovering Statistics Using SPSS (3rd ed.). London: Sage Publications. https://doi.org/10.14359/51686441
Flood, N. (2013). Seasonal composite landsat TM/ETM+ Images using the medoid (a multi-dimensional median). Remote Sensing, 5(12), 6481–6500. https://doi.org/10.3390/rs5126481
Forsyth, T. (2007). Sustainable livelihood approaches and soil erosion risks: Who is to judge? International Journal of Social Economics, 34(1–2), 88–102. https://doi.org/10.1108/03068290710723381
Franklin, S. E., & Wulder, M. A. (2002). Remote sensing methods in medium spatial resolution satellite data land cover classification of large areas. Progress in Physical Geography, 26(2), 173–205. https://doi.org/10.1191/0309133302pp332ra
García-Gutiérrez, C., Pachepsky, Y., & Martín, M. Á. (2018). Technical note: Saturated hydraulic conductivity and textural heterogeneity of soils. Hydrol. Earth Syst. Sci, 22, 3923–3932. https://doi.org/10.5194/hess-22-3923-2018
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
67
Gijsman, A. J., Jagtap, S. S., & Jones, J. W. (2003). Wading through a swamp of complete confusion : how to choose a method for estimating soil water retention parameters for crop models. European Journal of Agronomy, 18, 77–106. https://doi.org/10.1016/S1161-0301(02)00098-9
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., & Moore, R. (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202, 18–27. https://doi.org/10.1016/j.rse.2017.06.031
Green, W. H., & Ampt, G. A. (1911). Studies on Soil Physics. The Journal of Agricultural Science, 4(1), 1–24. https://doi.org/10.1017/S0021859600001441
Gregory, J. H., Dukes, M., Jones, P., & Miller, G. (2006). Effect of urban soil compaction on infiltration rate. Journal of Soil and Water Conservation, 61(3), 117–124.
Grunwald, S., Thompson, J. A., & Boettinger, J. L. (2011). Digital Soil Mapping and Modeling at Continental Scales: Finding Solutions for Global Issues. Soil Science Society of America Journal, 75(4), 1201–1213. https://doi.org/10.2136/sssaj2011.0025
Gu, D., & Gillespie, A. (1998). Topographic Normalization of Landsat TM Images of Forest Based on Subpixel Sun–Canopy–Sensor Geometry. Remote Sensing of Environment, 64(2), 166–175. https://doi.org/10.1016/S0034-4257(97)00177-6
Hamer, U., & Marschner, B. (2005). Priming effects in soils after combined and repeated substrate additions. Geoderma, 128(1–2), 38–51. https://doi.org/10.1016/J.GEODERMA.2004.12.014
Han, E., Ines, A., & Koo, J. (2015). Global High-Resolution Soil Profile Database for Crop Modeling Applications. Washington DC. Retrieved from https://harvestchoice.org/publications/global-high-resolution-soil-profile-database-crop-modeling-applications
Han, J., Kamber, M., & Pei, J. (2013). Getting to Know Your Data. In Data Mining Concepts and Techniques (3rd ed., pp. 66–79). Amsterdam: Elsevier. Retrieved from http://ir.obihiro.ac.jp/dspace/handle/10322/3933
Hansen, M. C., Egorov, A., Roy, D. P., Potapov, P., Ju, J., Turubanova, S., … Loveland, T. R. (2011). Continuous fields of land cover for the conterminous United States using Landsat data: first results from the Web-Enabled Landsat Data (WELD) project. Remote Sensing Letters, 2(4), 279–288. https://doi.org/10.1080/01431161.2010.519002
Hassler, S. K. (2013). Saturated hydraulic conductivity in the humid tropics : sources of variability, implications for monitoring and effects on near-surface hydrological flow paths. University of Potsdam. Retrieved from http://opus.kobv.de/ubp/volltexte/2013/6686/
Hassler, S. K., Lark, R. M., Milne, A. E., & Elsenbeer, H. (2011). Exploring the variation in soil saturated hydraulic conductivity under a tropical rainforest using the wavelet transform. European Journal of Soil Science, 62, 891–901. https://doi.org/10.1111/j.1365-2389.2011.01400.x
Hengl, T., Heuvelink, G. B. M., Kempen, B., Leenaars, J. G. B., Walsh, M. G., Shepherd, K. D., … Tondoh, J. E. (2015). Mapping Soil Properties of Africa at 250 m Resolution: Random Forests Significantly Improve Current Predictions. PLOS ONE, 10(6), 1–26. https://doi.org/10.1371/journal.pone.0125814
Hengl, T., Mendes De Jesus, J., Heuvelink, G. B. M., Gonzalez, M. R., Kilibarda, M., Blagotić, A., … Blagotić, M. (2017). SoilGrids250m: Global gridded soil information based on machine learning. PLOS ONE, 12(2), 1–40. https://doi.org/10.1371/journal
Hessel, R., Jetten, V., & Guanghui, Z. (2003). Estimating Manning’s n for steep slopes. CATENA, 54(1–2), 77–91. https://doi.org/10.1016/S0341-8162(03)00058-4
Hislop, S., Jones, S., Soto-Berelov, M., Skidmore, A., Haywood, A., & Nguyen, T. H. (2018). Using landsat spectral indices in time-series to assess wildfire disturbance and recovery. Remote Sensing, 10(3), 1–17. https://doi.org/10.3390/rs10030460
Horton, R. E. (1933). The Role of Infiltration in the Hydrologic Cycle. Transactions, American Geophysical Union, 446–460.
Housman, I., Tanpipat, V., Biswas, T., Clark, A., Stephen, P., Maus, P., & Megown, K. (2015). Monitoring Forest Change in Southeast Asia : Case Studies for USAID Lowering Emissions in Asia’s Forests. Retrieved from http://www.leafasia.org/sites/default/files/public/resources/USAIDLEAF-RSAC-ForestLossReport-November2015-Revised.pdf
Hudson, B. D. (1994). Soil organic matter and available water capacity. Journal of Soil and Water Conservation, 49(2), 189–194.
Huete, A. R. (1988). A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment, 25(3), 295–309. https://doi.org/10.1016/0034-4257(88)90106-X
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
68
Hussein, S. O., Kovács, F., & Tobak, Z. (2017). Spatiotemporal assessment of vegetation indices and land cover for Erbil City and its surrounding using MODIS imageries. Journal of Environmental Geography , 10(2), 31–39. https://doi.org/10.1515/jengeo-2017-0004
IFRC. (2016). World Disasters Report. International Federation of Red Cross and Red Crescent Societies (Vol. 91). https://doi.org/10.2307/2998097
IPCC. (2014). Summary for Policymakers. Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. https://doi.org/10.1017/CBO9781107415324
Jasiewicz, J., & Stepinski, T. F. (2013). Geomorphons — a pattern recognition approach to classification and mapping of landforms. Geomorphology, 182, 147–156. https://doi.org/10.1016/j.geomorph.2012.11.005
Jiang, Z., Huete, A. R., Didan, K., & Miura, T. (2008). Development of a two-band enhanced vegetation index without a blue band. Remote Sensing of Environment, 112(10), 3833–3845. https://doi.org/10.1016/J.RSE.2008.06.006
Kalyanapu, A. J., Burian, S. J., & Mcpherson, T. N. (2009). Effect of land use-based surface roughness on hydrologic model output. Journal of Spatial Hydrology, 9(2).
Khatami, R., Mountrakis, G., & Stehman, S. V. (2016). A meta-analysis of remote sensing research on supervised pixel-based land-cover image classification processes: General guidelines for practitioners and future research. Remote Sensing of Environment, 177(May), 89–100. https://doi.org/10.1016/j.rse.2016.02.028
Kim, C. (2016). Land use classification and land use change analysis using satellite images in Lombok Island, Indonesia. Forest Science and Technology, 12(4), 183–191. https://doi.org/10.1080/21580103.2016.1147498
Kizilkaya, R., & Dengiz, O. (2010). Variation of land use and land cover effects on some soil physico-chemical characteristics and soil enzyme activity. Zemdirbyste-Agriculture, 97(2), 15–24. Retrieved from https://pdfs.semanticscholar.org/2f5b/b343e51e503b68b6675b6c7a40a0fd8ddd1b.pdf
Koppad, A. G., & Janagoudar, B. S. (2013). Vegetation analysis and land use land cover classification of forest in Uttara Kannada district India using remote sensign and GIS techniques. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives, 42(4W5), 121–125. https://doi.org/10.5194/isprs-archives-XLII-4-W5-121-2017
Kuriakose, S. L., Devkota, S., Rossiter, D. G., & Jetten, V. G. (2009). Prediction of soil depth using environmental variables in an anthropogenic landscape, a case study in the Western Ghats of Kerala, India. Catena, 79(1), 27–38. https://doi.org/10.1016/j.catena.2009.05.005
Lado, M., Paz, A., & Ben-Hur, M. (2004). Organic Matter and Aggregate-Size Interactions in Saturated Hydraulic Conductivity. Soil Science Society of America Journal, 68(1), 234–242. https://doi.org/10.2136/sssaj2004.2340
Lemmon, P. E. (2008). Using Forest Densiometers - Documentation. Jackson: Forestry Suppliers, Inc. Liaw, A., & Wiener, M. (2002). Classification and Regression by randomForest. R News, 2(3), 18–22.
https://doi.org/10.1159/000323281 Lin, Y., & Wei, X. (2008). The impact of large-scale forest harvesting on hydrology in the Willow
watershed of Central British Columbia. Journal of Hydrology, 359(1–2), 141–149. https://doi.org/10.1016/j.jhydrol.2008.06.023
Looy, K. Van, Bouma, J., Herbst, M., Koestel, J., Minasny, B., Mishra, U., … Looy, V. (2017). Pedotransfer Functions in Earth System Science: Challenges and Perspectives. AGU Publications. https://doi.org/10.1002/2017RG000581
Lubis, J. P. G., & Nakagoshi, N. (2011). Land Use and Land Cover Change Detection using Remote Sensing and Geographic Information System in Bodri Watershed , Central Java , Indonesia. Journal of International Development and Cooperation, 18(1), 139–151.
Mallupattu, P. K., & Reddy, J. S. R. (2013). Analysis of Land Use/Land Cover Changes Using Remote Sensing Data and GIS at an Urban Area, Tirupati, India. The Scientific World Journal, 2013, 1–6. https://doi.org/10.1155/2013/268623
Malo, D. D., Worcester, B. K., Cassel, D. K., & Matzdorf, K. D. (1974). Soil-Landscape Relationships in a Closed Drainage System. Soil Science Society of America Journal, 38(5), 813–818. https://doi.org/10.2136/sssaj1974.385813x
Marchi, L. (2017). Linking Debris Flows and Landslides to Large Floods. In D. Tsutsumi & J. B. Laronne (Eds.), Gravel-bed rivers : processes and disasters (1st ed., pp. 467–494). John Wiley & Sons Ltd.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
69
Marchi, L., Borga, M., Preciso, E., & Gaume, E. (2010). Characterisation of selected extreme flash floods in Europe and implications for flood risk management. Journal of Hydrology, 394(1–2), 118–133. https://doi.org/10.1016/j.jhydrol.2010.07.017
Mas, J.-F. (1999). Monitoring land-cover changes: a comparison of change detection techniques*. International Journal of Remote Sensing, 20(1), 139–152. Retrieved from http://www.tandf.co.uk/JNLS/res.htm
Maselli, F., Conese, C., De Filippis, T., & Romani, M. (1995). Integration of ancillary data into a maximum-likelihood classifier with nonparametric priors. ISPRS Journal of Photogrammetry and Remote Sensing, 50(2), 2–11. https://doi.org/10.1016/0924-2716(95)98210-Q
McBratney, A. B., Mendonça Santos, M. L., & Minasny, B. (2003). On digital soil mapping. Geoderma, 117(1–2), 3–52. https://doi.org/10.1016/S0016-7061(03)00223-4
McFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing, 17(7), 1425–1432. https://doi.org/10.1080/01431169608948714
McKenzie, N. J., Grundy, M., Webster, R., & Ringrose-Voase, A. J. (2008). Guidlines for Surveying Soil and Land Resources (2nd ed.). Melbourne: CSIRO Publishing.
Merritt, W. S., Letcher, R. A., & Jakeman, A. J. (2003). A review of erosion and sediment transport models. Environmental Modelling & Software, 18, 761–799. https://doi.org/10.1016/S1364-8152(03)00078-1
Miller, B. A., & Schaetzl, R. J. (2015). Digital Classification of Hillslope Position. Soil Science Society of America Journal, 79(1), 132–145. https://doi.org/10.2136/sssaj2014.07.0287
Möller, M., Volk, M., Friedrich, K., & Lymburner, L. (2008). Placing soil-genesis and transport processes into a landscape context: A multiscale terrain-analysis approach. Journal of Plant Nutrition and Soil Science, 171(3), 419–430. https://doi.org/10.1002/jpln.200625039
Morgan, R. P. C., Quinton, J. N., Smith, R. E., Govers, G., Prosen, J. W. A., Auerswald, K., … Folley, A. J. V. (1998). The European Soil Erosion Model (EUROSEM): documentation and user guide.
Muñoz, C., Zagal, E., & Ovalle, C. (2007). Influence of trees on soil organic matter in Mediterranean agroforestry systems: an example from the “Espinal” of central Chile. European Journal of Soil Science, 58(3), 728–735. https://doi.org/10.1111/j.1365-2389.2006.00858.x
Nemes, A., Rawls, W. J., & Pachepsky, Y. A. (2005). Influence of Organic Matter on the Estimation of Saturated Hydraulic Conductivity. Soil Science Society of America Journal, 69(4), 1330. https://doi.org/10.2136/sssaj2004.0055
Nguyen, L. H., Joshi, D. R., Clay, D. E., & Henebry, G. M. (2018). Characterizing land cover/land use from multiple years of Landsat and MODIS time series: A novel approach using land surface phenology modeling and random forest classifier. Remote Sensing of Environment. https://doi.org/10.1016/J.RSE.2018.12.016
Nyberg, G., Bargués Tobella, A., Kinyangi, J., & Ilstedt, U. (2012). Soil property changes over a 120-yr chronosequence from forest to agriculture in western Kenya. Hydrology and Earth System Sciences, 16(7), 2085–2094. https://doi.org/10.5194/hess-16-2085-2012
Ohm, H., Hamer, U., & Marschner, B. (2007). Priming effects in soil size fractions of a podzol Bs horizon after addition of fructose and alanine. Journal of Plant Nutrition and Soil Science, 170(4), 551–559. https://doi.org/10.1002/jpln.200625087
Ossola, A., Hahs, A. K., & Livesley, S. J. (2015). Habitat complexity influences fine scale hydrological processes and the incidence of stormwater runoff in managed urban ecosystems. Journal of Environmental Management, 159, 1–10. https://doi.org/10.1016/j.jenvman.2015.05.002
Panahi, A., Alijani, B., & Mohammadi, H. (2010). The Effect of the Land Use/Cover Changes on the Floods of the Madarsu Basin of Northeastern Iran. Journal of Water Resource and Protection, 02(04), 373–379. https://doi.org/10.4236/jwarp.2010.24043
Paramananthan, S., Lee, P.-X., Wong, M.-K., Van Ranst, E., Wüst, R. A. J., & Vijiandran, J. R. (2018). A Comparative Study of the Use of Organic Carbon and Loss on Ignition in Defining Tropical Organic Soil Materials. Communications in Soil Science and Plant Analysis, 49(5), 626–634. https://doi.org/10.1080/00103624.2018.1435683
Pekel, J.-F., Cottam, A., Gorelick, N., & Belward, A. S. (2016). High-resolution mapping of global surface water and its long-term changes. Nature, 540(7633), 418–422. https://doi.org/10.1038/nature20584
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
70
Peng, S. L., Wu, J., & You, W. H. (2012). Recovery of saturated hydraulic conductivity along a forest successional series from abandoned land to mature, evergreen broad-leaved forest in eastern China. Soil Research, 50, 257–266. Retrieved from http://tiantong.ecnu.edu.cn/ueditor/net/upload/2014-09-19/035e24eb-4005-4179-a321-de1f616db2a8.pdf
Pérez-Molina, E., Sliuzas, R., Flacke, J., & Jetten, V. (2017). Developing a cellular automata model of urban growth to inform spatial policy for flood mitigation: A case study in Kampala, Uganda. Computers, Environment and Urban Systems, 65, 53–65. https://doi.org/10.1016/j.compenvurbsys.2017.04.013
Powell, S. L., Cohen, W. B., Healey, S. P., Kennedy, R. E., Moisen, G. G., Pierce, K. B., & Ohmann, J. L. (2009). Quantification of live aboveground forest biomass dynamics with Landsat time-series and field inventory data: A comparison of empirical modeling approaches. Remote Sensing of Environment, 114, 1053–1068. https://doi.org/10.1016/j.rse.2009.12.018
Pramono, A. ., Fauzi, M. A., Widyani, N., Heriansyah, I., & Roshetko, J. M. (2015). Managing smallholder teak plantations: Field guide for farmers (1st ed.). Bogor: Center for International Forestry Research (CIFOR). https://doi.org/10.17528/cifor/003493
R Core Team. (2018). A Language and Environment for Statistical Computing. Retrieved June 17, 2019, from https://www.r-project.org/
Raudkivi, A. J. (1979). Hydrology : an advanced introduction to hydrological processes and modelling (1st ed.). Oxford: Pergamon Press.
Riaño, D., Chuvieco, E., Salas, J., & Aguado, I. (2003). Assessment of Different Topographic Corrections in Landsat-TM Data for Mapping Vegetation Types. IEEE Transactions on Geoscience and Remote Sensing, 41(5). https://doi.org/10.1109/TGRS.2003.811693
Rittima, A., Piemfa, K., Uthai, N., & Jantaramana, A. (2013). Improvement of Design Rainfall Analysis of the Central Basin of Thailand. Engineering Research and Development, 24(4), 28–38.
Rodriguez-Galiano, V. F., Ghimire, B., Rogan, J., Chica-Olmo, M., & Rigol-Sanchez, J. P. (2012). An assessment of the effectiveness of a random forest classifier for land-cover classification. ISPRS Journal of Photogrammetry and Remote Sensing, 67(1), 93–104. https://doi.org/10.1016/j.isprsjprs.2011.11.002
Ross, S. M. (1993). Organic matter in tropical soils: current conditions, concerns and prospects for conservation. Progress in Physical Geography: Earth and Environment, 17(3), 265–305. https://doi.org/10.1177/030913339301700301
Roy, D.P., Zhang, H. K., Ju, J., Gomez-Dans, J. L., Lewis, P. E., Schaaf, C. B., … Kovalskyy, V. (2016). A general method to normalize Landsat reflectance data to nadir BRDF adjusted reflectance. Remote Sensing of Environment, 176, 255–271. https://doi.org/10.1016/J.RSE.2016.01.023
Roy, David P., Li, J., Zhang, H. K., Yan, L., Huang, H., & Li, Z. (2017). Examination of Sentinel-2A multi-spectral instrument (MSI) reflectance anisotropy and the suitability of a general method to normalize MSI reflectance to nadir BRDF adjusted reflectance. Remote Sensing of Environment, 199, 25–38. https://doi.org/10.1016/J.RSE.2017.06.019
Rwanga, S. S., & Ndambuki, J. M. (2017). Accuracy Assessment of Land Use / Land Cover Classification Using Remote Sensing and GIS. International Journal of Geosciences, (8), 611–622. https://doi.org/10.4236/ijg.2017.84033
Saah, D., Tenneson, K., Poortinga, A., Nguyen, Q., Chishtie, F., & San Aung, K. (2019). A novel modular free and open-source system for landcover mapping using cloud-based remote sensing and machine learning. Remote Sensing of Environment, in review.
Sajikumar, N., & Remya, R. S. (2015). Impact of land cover and land use change on runoff characteristics. Journal of Environmental Management, 161, 460–468. https://doi.org/10.1016/j.jenvman.2014.12.041
Salomonson, V. ., & Appel, I. (2004). Estimating fractional snow cover from MODIS using the normalized difference snow index. Remote Sensing of Environment, 89(3), 351–360. https://doi.org/10.1016/J.RSE.2003.10.016
Sanchez-Moreno, J. F., Jetten, V., Mannaerts, C. M., & Tavares, J. P. (2014). Selecting best mapping strategies for storm runoff modeling in a mountainous semi-arid area. Earth Surface Processes and Landforms, 39(8), 1030–1048. https://doi.org/10.1002/esp.3501
Sanchez, P. A., Ahamed, S., Carre, F., Hartemink, A. E., Hempel, J., Huising, J., … Zhang, G.-L. (2009). Digital Soil Map of the World. Science, 325(5941), 680–681. https://doi.org/10.1126/science.1175084
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
71
Saxton, K. E., & Rawls, W. J. (2006). Soil Water Characteristic Estimates by Texture and Organic Matter for Hydrologic Solutions. Soil Science Society of America Journal, 70, 1569–1578. https://doi.org/10.2136/sssaj2005.0117
Saxton, K. E., Rawls, W. J., Romberger, J. S., & Papendick, R. I. (1986). Estimating Generalized Soil-water Characteristics from Texture. Soil Science Society of America Journal, 50(4), 1031–1036. https://doi.org/10.2136/sssaj1986.03615995005000040039x
Schaetzl, R. J., & Anderson, S. (2005). Soils Genesis and Geomorphology (1st ed.). New York: Cambridge University Press. https://doi.org/10.2136/sssaj2014.07.0287
Scharffenberg, W. (2016). Hydrologic Modeling System HEC-HMS: User’s Manual. Davis. Retrieved from http://www.hec.usace.army.mil/software/hec-hms/documentation/HEC-HMS_Users_Manual_4.2.pdf
Schmidt, G., Jenkerson, C., Masek, J., Vermote, E., & Gao, F. (2013). Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) Algorithm Description, Technical Report. Reston. Retrieved from http://www.usgs.gov/pubprod
Schoeneberger, P. J., Wysocki, D. A., & Benham, E. C. (2012). Field Book for Describing and Sampling Soils (3rd ed.). Lincoln: National Resources Conservation Service, National Soil Survey Center. Retrieved from https://www.nrcs.usda.gov/Internet/FSE_DOCUMENTS/nrcs142p2_052523.pdf
Schulte, E. E., & Hopkins, B. G. (1996). Estimation of Soil Organic Matter by Weight Loss-On-Ignition. In Soil Organic Matter: Analysis and Interpretation (pp. 21–31). Soil Science Society of America. https://doi.org/10.2136/sssaspecpub46.c3
Shekhar, S., & Xiong, H. (2008). Soil and Water Assessment Tool “SWAT.” In Encyclopedia of GIS. Boston, MA: Springer US. https://doi.org/10.1007/978-0-387-35973-1_1231
Shrestha, D. P. (2014). INCREO project report - rainfall -runoff modelling in the Fella river basin. Enschede. Sidle, R. C., Ziegler, A. D., Negishi, J. N., Nik, A. R., Siew, R., & Turkelboom, F. (2006). Erosion
processes in steep terrain—Truths, myths, and uncertainties related to forest management in Southeast Asia. Forest Ecology and Management, 224(1–2), 199–225. https://doi.org/10.1016/J.FORECO.2005.12.019
Sliuzas, R., Flacke, J., & Jetten, V. (2013). Modelling Urbanization and Flooding in Kampala, Uganda. In Network-Association of European Researchers on Urbanisation in the South (N-AERUS) XIV (pp. 12–14). Enschede. Retrieved from https://pdfs.semanticscholar.org/546f/2ea13e914dc89c141eff4816312ea5198186.pdf
Sluiter, R., & Pebesma, E. J. (2010). Comparing techniques for vegetation classification using multi- and hyperspectral images and ancillary environmental data. International Journal of Remote Sensing, 31(23), 6143–6161. https://doi.org/10.1080/01431160903401379
Smith, J. A., Lin, T. L., & Ranson, K. J. (1980). The Lambertian Assumption and Landsat Data. Photogrammetric Engineering and Remote Sensing, 46(9), 1183–1189. Retrieved from https://www.asprs.org/wp-content/uploads/pers/1980journal/sep/1980_sep_1183-1189.pdf
Sobieraj, J. A., Elsenbeer, H., Coelho, R. M., & Newton, B. (2002). Spatial variability of soil hydraulic conductivity along a tropical rainforest catena. Geoderma, 108(1–2), 79–90. https://doi.org/10.1016/S0016-7061(02)00122-2
Soenen, S. A., Peddle, D. R., & Coburn, C. A. (2005). SCS+C: a modified Sun-canopy-sensor topographic correction in forested terrain. IEEE Transactions on Geoscience and Remote Sensing, 43(9), 2148–2159. https://doi.org/10.1109/TGRS.2005.852480
Soil Survey Staff. (2014). Soil Survey Field and Laboratory Methods Manual. Steinhausen, M. J., Wagner, P. D., Narasimhan, B., & Waske, B. (2018). Combining Sentinel-1 and
Sentinel-2 data for improved land use and land cover mapping of monsoon regions. International Journal of Applied Earth Observation and Geoinformation, 73(April), 595–604. https://doi.org/10.1016/j.jag.2018.08.011
Stepinski, T. F., & Jasiewicz, J. (2011). Geomorphons - a new approach to classification of landforms. Geomorphometry, 109–112.
Tifafi, M., Guenet, B., & Hatté, C. (2018). Large Differences in Global and Regional Total Soil Carbon Stock Estimates Based on SoilGrids, HWSD, and NCSCD: Intercomparison and Evaluation Based on Field Data From USA, England, Wales, and France. Global Biogeochemical Cycles, 32(1), 42–56. https://doi.org/10.1002/2017GB005678
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
72
Tilaar, M., Wih, W. L., Ranti, A. S., Wasitaatmadja, S. M., Junardy, F. D., & Maily. (2008). Review of Lansium domesticum Corrêa and its use in cosmetics. Boletin Latinoamericano y Caribe de Plantas Medicinales y Aromáticas, 7(4), 183–189. Retrieved from https://pdfs.semanticscholar.org/a844/b6f156e3db3a12ded31f65c5e63ef9c70d11.pdf
Tsai, Y. H., Stow, D., Chen, H. L., Lewison, R., An, L., & Shi, L. (2018). Mapping vegetation and land use types in Fanjingshan National Nature Reserve using google earth engine. Remote Sensing, 10(6). https://doi.org/10.3390/rs10060927
USAID. (2015). Burma Floods Fact Sheet. Retrieved from https://www.usaid.gov/sites/default/files/documents/1866/burma_fl_fs01_08-06-2015.pdf
Usamah, M., & Arambepola, N. (2013). Lessons Learned from the 2006 Flashfloods and Landslide in Uttaradit and Sukhothai Provinces: Implication for Effective Landslide Disaster Risk Mangement in Thailand. Landslide Science and Practice, 6(May 2011), 693–699. https://doi.org/10.1007/978-3-642-31310-3
Van Westen, C., Jetten, V., Trigg, M., & Alkema, D. (2015). Report on methodologies for hazard mapping in the Caribbean. Retrieved from http://www.charim.net/sites/default/files/handbook/otherpages/CHARIM Methodology Report.pdf
Vegetation Subcommittee. (1997). Vegetation Classification Standard. Reston. Retrieved from http://www.fgdc.gov/fgdc.html
Vermote, E., Justice, C., Claverie, M., & Franch, B. (2016). Preliminary analysis of the performance of the Landsat 8/OLI land surface reflectance product. Remote Sensing of Environment, 185, 46–56. https://doi.org/10.1016/J.RSE.2016.04.008
Veselsky, M., Bandura, P., Burian, L., Harcinikova, T., & Bella, P. (2015). Semi-automated recognition of planation surfaces and other flat landforms: A case study from the Aggtelek Karst, Hungary. Open Geosciences, 7(1), 799–811. https://doi.org/10.1515/geo-2015-0063
Wardlow, B. D., Egbert, S. L., & Kastens, J. H. (2007). Analysis of Time-Series MODIS 250 m Vegetation Index Data for Crop Classification in the U.S. Central Great Plains "Analysis of Time-Series MODIS 250 m Vegetation Index Data for Crop Classification in the U.S. Central Great Plains. Remote Sensing of Environment, 108(3), 290–310. https://doi.org/10.1016/j.rse.2006.11.021
Wijaya, K., Nishimura, T., Setiawan, B. I., & Saptomo, S. K. (2009). Spatial variability of soil saturated hydraulic conductivity in paddy field in accordance to subsurface percolation. Paddy and Water Environment, 8(2), 113–120. https://doi.org/10.1007/s10333-009-0190-x
Williams, R. D., Ahuja, L. R., & Naney, J. W. (1992). Comparison of Methods to Estimate Soil Water Characteristics from Soil Texture, Bulk Density, and Limited Data. Soil Science, 153(3), 172–184. Retrieved from https://insights-ovid-com.ezproxy2.utwente.nl/crossref?an=00010694-199203000-00002
Wösten, J. H. M., Lilly, A., Nemes, A., & Le Bas, C. (1999). Development and use of a database of hydraulic properties of European soils. Geoderma, 90(3–4), 169–185. https://doi.org/10.1016/S0016-7061(98)00132-3
Xu, H. (2008). A new index for delineating built-up land features in satellite imagery. International Journal of Remote Sensing, 29(14), 4269–4276. https://doi.org/10.1080/01431160802039957
Xu, Hanqiu. (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing, 27(14), 3025–3033. https://doi.org/10.1080/01431160600589179
Yen, B. C., & Chow, V. T. (1980). Design Hyetographs for Small Drainage Structures. Journal of the Hydraulic Division, 1055–1076.
Young, N. E., Anderson, R. S., Chignell, S. M., Vorster, A. G., Lawrence, R., & Evangelista, P. H. (2017). A survival guide to Landsat preprocessing. Ecology, 98(4), 920–932. https://doi.org/10.1002/ecy.1730
Zha, Y., Gao, J., & Ni, S. (2003). Use of normalized difference built-up index in automatically mapping urban areas from TM imagery. International Journal of Remote Sensing, 24(3), 583–594. https://doi.org/10.1080/01431160210144570
Zhu, Z., Gallant, A. L., Woodcock, C. E., Pengra, B., Olofsson, P., Loveland, T. R., … Auch, R. F. (2016). Optimizing selection of training and auxiliary data for operational land cover classification for the LCMAP initiative. ISPRS Journal of Photogrammetry and Remote Sensing, 122, 206–221. https://doi.org/10.1016/J.ISPRSJPRS.2016.11.004
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
73
Zhu, Z., & Woodcock, C. E. (2011). Object-based cloud and cloud shadow detection in Landsat imagery. Remote Sensing of Environment, 118, 83–94. https://doi.org/10.1016/j.rse.2011.10.028
Zimmermann, B., Elsenbeer, H., & De Moraes, J. M. (2006). The influence of land-use changes on soil hydraulic properties: Implications for runoff generation. Forest Ecology and Management, 222(1–3), 29–38. https://doi.org/10.1016/J.FORECO.2005.10.070
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
74
ANNEXES
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
75
Annex 1. Geological map of Uttaradit (Ministry of Natural Resources and Environment Thailand, 2019)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
76
Annex 2. Pedotransfer script by Jetten and Shrestha (2018) based on the equations of Saxton et al. (2006) # Model: Saxtons pedotransfer function SWAP model 2005 # # # input data SOILGRIDS.ORG # # Date: 05/04/2018 # # Version: 1.1 # # Author: V Jetten @ ITC # ######################################################## # $1 = soilgrids layer indication, e.g. "sl2" # $2 is lisem layer, 1 or 2 # $3 is the degree of saturation between porosity and field capacity, # used for the initial moisture content and initial suction head # $4 is the bulk density you consider normal (uncompacted and not loose) in the area in kg/m3 #! ;; sl2 1 0.7 binding S = SNDPPT_M_sl2_250m.tif; #sand % C = CLYPPT_M_sl2_250m.tif; #clay % # OC = oc$1.map; #organic carbon in % Gravel = CRFVOL_M_sl2_250m.tif; #coarse fragments %, note in excel sheet it says g/cc # but this is not correct, it is used as a volume fraction OC = ORCDRC_M_sl2_250m.tif; bdsg=BLDFIE_M_sl2_250m.tif; #bulkdensity kg/m3 standardBD = scalar(1470); # <= used to calibrate output 1470 fractionmoisture = scalar(0.7); #inital moisture as fraction between porosity and field capacity # 0 = init moist is at FC, 1.0 = init moist is at porosity POROSITY = pore.map; #porosity (cm3/cm3) Ksat = ksat.map; # ksat in mm/h initmoist =thetainit.map; # inital moisture (cm3/cm3) psi=psi1.map; # suction with init moisture in cm, used in LISEM se = se.map; # relative moisture content between 0-1 Densityfactor = densfact.map; BD = bulkdens.map; # ton/m3 WP = wilting.map; # wilting point moisture content FC = fieldcap.map; # field capacity moisture content PAW = plantAVW.map; sand = sand.map; clay = clay.map; grav = graveln.map; dem = dem30.map; initial # prep data S = S/100; C = C/100; OC= (OC/1000)*100; # conversion OC from g/kg to percentage OM = OC*1.72; #conversion org carbon to org matter report om.map = OM; Gravel = Gravel/100; report Densityfactor = bdsg/standardBD;#scalar(1.0); # upper boundary 1.15 standardBD # calculated as the bulk density from soilgrids divided by some standard bd # multiple regression from excel # wilting point stuff M1500 =-0.024*S+0.487*C+0.006*OM+0.005*S*OM-0.013*C*OM+0.068*S*C+0.031; #W18) # =-0.024*F18+0.487*G18+0.006*H18+0.005*F18*H18-0.013*G18*H18+0.068*F18*G18+0.031 M1500adj =M1500+0.14*M1500-0.02; #X18) =W18+0.14*W18-0.02
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
77
# field capacity stuff M33 =-0.251*S+0.195*C+0.011*OM+0.006*S*OM-0.027*C*OM+0.452*S*C+0.299; #Y18) #=-0.251*F18+0.195*G18+0.011*H18+0.006*F18*H18-0.027*G18*H18+0.452*F18*G18+0.299 M33adj = M33+(1.283*M33*M33-0.374*M33-0.015); #Z18) =Y18+(1.283*Y18*Y18-0.374*Y18-0.015) # porosity - FC PM33 = 0.278*S+0.034*C+0.022*OM-0.018*S*OM-0.027*C*OM-0.584*S*C+0.078; #AA18) #=0.278*F18+0.034*G18+0.022*H18-0.018*F18*H18-0.027*G18*H18-0.584*F18*G18+0.078 PM33adj = PM33+(0.636*PM33-0.107); #AB18) =AA18+(0.636*AA18-0.107) # porosity SatPM33 = M33adj + PM33adj; #AC18) =AB18+Z18 SatSadj = -0.097*S+0.043; #AD18) =-0.097*F18+0.043 SadjSat = SatPM33 + SatSadj; #AE18) =AC18+AD18 Dens_om = (1-SadjSat)*2.65; #AF18) =(1-AE18)*2.65 Dens_comp = Dens_om * Densityfactor; #AG18) =AF18*(I18) PORE_comp =(1-Dens_om/2.65)-(1-Dens_comp/2.65); #AI18) =(1-AG18/2.65)-(1-AF18/2.65) M33comp = M33adj + 0.2*PORE_comp; #AJ18) =Z18+0.2*AI18 #output report POROSITY = 1-(Dens_comp/2.65); #AH18) PoreMcomp = POROSITY-M33comp; #AK18) LAMBDA = (ln(M33comp)-ln(M1500adj))/(ln(1500)-ln(33)); #AL18) GravelRedKsat =(1-Gravel)/(1-Gravel*(1-1.5*(Dens_comp/2.65))); #AM18) report Ksat = max(0.0, 1930*(PoreMcomp)**(3-LAMBDA)*GravelRedKsat); #AN18) report BD = Gravel*2.65+(1-Gravel)*Dens_comp; #U18 report WP = M1500adj; report FC = M33adj; report PAW = (M33adj - M1500adj)*(1-Gravel); report initmoist= fractionmoisture*POROSITY+ (1-fractionmoisture)*FC; # A = exp[ln(33) + B ln(T33)] # B = [ln(1500) - ln(33)] / [ln(T33) - ln(T1500)] bB = (ln(1500)-ln(33))/(ln(FC)-ln(WP)); aA = exp(ln(33)+bB*ln(FC)); report psi= aA * initmoist**-bB *100/9.8; report se.map = initmoist/POROSITY;
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
78
Annex 2.1. Soil sampling locations Ban Da Na Kham watershed
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
79
Annex 2.2. Morphological properties of the soil for each site (field measurements)
Site Longitude Latitude Altitude
Soil Depth
(cm) Soil Colour
Clay
(%)
Sand
(%)
Silt
(%)
Soil Textural
Class
1 619771 1967551 204 80 Reddish Grey 22 36 42 Loam 2 618856 1968672 168 90 Very Pale Brown 17 62 22 Sandy Loam 3 618942 1968680 158 - Brown 27 40 33 Sandy Loam 4 618249 1967020 123 100 Light Brown 10 63 27 Sandy Loam 5 618225 1969653 154 100 Very Pale Brown 37 19 44 Loam 6 618984 1970672 251 80 Reddish Yellow 24 25 51 Loam 7 617962 1971016 170 100 Pinkish Grey 22 40 37 Silty Loam 8 617491 1969031 154 100 Yellow 19 40 41 Loam 9 617105 1968991 206 100 Brownish Yellow 42 14 44 Sandy Loam
10 617240 1969081 168 100 Light Grey 14 57 28 Clay Loam 11 616160 1966187 129 100 Light Yellowish Brown 2 60 38 Sandy Loam 12 614673 1967368 206 90 Light Yellowish Brown 24 27 49 Sandy Loam 13 614773 1967203 144 100 Pale Brown 23 37 40 Sandy Loam 14 612130 1969421 163 100 Brownish Yellow 31 18 51 Clay Loam 15 612205 1969377 168 100 Very Pale Brown 20 43 37 Clay Loam 16 616224 1966862 119 100 Light Grey 42 20 38 Sandy Loam 17 616212 1966923 137 100 Very Pale Brown 25 34 41 Silty Loam 18 616236 1966966 150 100 Very Pale Brown 11 50 38 Loam 19 610112 1970044 227 100 Yellow 31 20 49 Silty Loam 20 610574 1972634 228 100 Pale Brown 20 37 43 Sandy Loam 21 610571 1972681 230 100 Very Pale Brown 27 23 50 Sandy Loam 22 610531 1972616 214 80 Light Grey 20 42 38 Loam 23 609624 1972236 278 100 Brownish Yellow 28 29 43 Clay Loam 24 609712 1972273 292 100 Light Yellowish Brown 23 31 47 Loam 25 609486 1972244 224 60 Pale Brown 16 54 30 Clay Loam 26 610540 1972106 200 100 Light Yellowish Brown 19 29 52 Sandy Loam 27 610484 1972051 220 100 Very Pale Brown 27 15 59 Silty Loam 28 610577 1971984 259 100 Very Pale Brown 32 21 47 Loam 29 610219 1969983 211 62 Light Yellowish Brown 17 43 40 Loam 30 612991 1969365 139 100 Light Yellowish Brown 12 48 40 Loam 31 615578 1970494 175 100 Very Pale Brown 29 34 38 Loam 32 615625 1970421 207 100 Brownish Yellow 38 23 39 Silty Loam 33 615552 1970369 180 61 Very Pale Brown 37 23 40 Clay Loam 34 615229 1971325 169 25 Very Pale Brown 14 41 45 Loam 35 615286 1971318 177 100 Yellow 30 32 38 Loam 36 619063 1969265 162 100 Very Pale Brown 30 24 46 Clay Loam 37 619363 1972050 226 90 Greyish Brown 18 41 41 Loam 38 619368 1972261 162 100 Brown 11 59 30 Loam 39 618396 1972508 255 100 Yellow 2 61 37 Loam 40 618421 1972541 238 100 Yellow 6 54 40 Silty Clay 41 618471 1972653 217 100 Light Yellowish Brown 16 40 44 Clay Loam 42 612490 1972619 380 100 Light Yellowish Brown 5 45 50 Loam 43 612532 1972193 307 100 Pale Brown 2 58 40 Loam 44 610959 1972270 232 100 Light Yellowish Brown 2 53 45 Clay 45 611772 1970705 174 100 Light Yellowish Brown 2 49 49 Clay Loam 46 608972 1972152 233 75 Pale Brown 10 45 45 Loam 47 611828 1970614 160 100 Very Pale Brown 2 45 53 Sandy Loam 48 616782 1965176 110 100 Light Grey 4 58 38 Silty Loam
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
80
Annex 2.3. Physical and chemical properties of the soil for each site (field measurements)
Site Longitude Latitude
Slope
Position Land Cover
Ks
(mm h-1)
Porosity
(%)
Bulk Density
(g cm-3) OM (%) 1 619771 1967551 Backslope Orchard 306 50 1.32 3.7
2 618856 1968672 Summit Cropland 2444 47 1.39 2.8
3 618942 1968680 Valley Cropland 87 35 1.72 2.4 4 618249 1967020 Valley Cropland 41 38 1.64 2.9 5 618225 1969653 Backslope Orchard 81 51 1.29 2.2 6 618984 1970672 Backslope Cropland 382 54 1.21 2.8 7 617962 1971016 Valley Cropland 68 46 1.42 2.0 8 617491 1969031 Valley Mixed Forest 417 55 1.20 2.8
9 617105 1968991 Summit Mixed Forest 3703 70 0.80 6.2
10 617240 1969081 Valley Orchard 978 45 1.45 2.9 11 616160 1966187 Valley Teak Plantation 468 46 1.43 3.2 12 614673 1967368 Summit Mixed Forest 191 60 1.07 4.8 13 614773 1967203 Backslope Orchard 917 63 0.99 5.9
14 612130 1969421 Backslope Mixed Forest 2657 55 1.20 5.6
15 612205 1969377 Valley Cropland 560 54 1.22 4.2 16 616224 1966862 Valley Mixed Forest 815 60 1.06 3.3
17 616212 1966923 Backslope Mixed Forest 2078 63 0.98 4.6
18 616236 1966966 Backslope Mixed Forest 204 50 1.32 5.1 19 610112 1970044 Backslope Orchard 1069 59 1.09 3.5 20 610574 1972634 Backslope Orchard 59 51 1.30 4.0 21 610571 1972681 Backslope Orchard 905 60 1.07 3.7 22 610531 1972616 Valley Orchard 1782 59 1.08 5.3 23 609624 1972236 Backslope Orchard 436 56 1.18 3.8 24 609712 1972273 Summit Orchard 1461 53 1.24 6.1 25 609486 1972244 Valley Orchard 417 51 1.29 5.7 26 610540 1972106 Valley Orchard 12 49 1.36 5.8 27 610484 1972051 Backslope Orchard 322 59 1.09 6.7 28 610577 1971984 Summit Orchard 382 61 1.03 3.9 29 610219 1969983 Valley Orchard 7 52 1.27 4.7 30 612991 1969365 Valley Orchard 726 54 1.22 3.6 31 615578 1970494 Backslope Orchard 92 58 1.12 5.1 32 615625 1970421 Backslope Mixed Forest 1986 76 0.64 8.8 33 615552 1970369 Valley Orchard 62 59 1.07 4.9 34 615229 1971325 Valley Orchard 229 48 1.39 3.6 35 615286 1971318 Backslope Orchard 664 52 1.27 5.2 36 619063 1969265 Backslope Orchard 5 41 1.58 2.7 37 619363 1972050 Valley Orchard 153 48 1.37 6.0
38 619368 1972261 Backslope Orchard 10784 64 0.95 6.5
39 618396 1972508 Summit Orchard 244 57 1.15 4.7 40 618421 1972541 Backslope Orchard 643 61 1.03 4.1 41 618471 1972653 Valley Orchard 323 61 1.04 4.8 42 612490 1972619 Backslope Teak Plantation 719 55 1.19 6.1 43 612532 1972193 Backslope Orchard 0.1 54 1.21 4.6 44 610959 1972270 Backslope Orchard 85 55 1.20 6.9 45 611772 1970705 Backslope Orchard 218 63 0.97 6.4 46 608972 1972152 Backslope Orchard 4 56 1.17 6.9 47 611828 1970614 Valley Orchard 17 50 1.32 3.8 48 616782 1965176 Backslope Orchard 16 43 1.50 3.9
Ks measurements framed in red were excluded from the soil analysis
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
81
Annex 2.4. Description of laboratory measurements for soil property determination
Saturated Hydraulic Conductivity
Ks was measured using the Constant Head method as described in the operating instructions for a
laboratory-permeameter (Eijkelkamp, 2013). Since the laboratory equipment did not include a permeameter,
an alternative apparatus had to be constructed. Before the measurements the samples were first soaked in
water for 24 hours to reach the saturated state. To prevent disturbance the samples were covered on one
site with a thin nylon membrane. Further, a plastic tape was attached to the upper half of the sampling ring
so that around 3 – 4 cm of the tape protruded. The sample was then placed in the funnel and the protruding
tape filled with water. During the measurement the water level above the sample was kept constant in order
to have no changes in pressure. The measurement was considered as final when the recorded amount of
water flowing through the sample reached a constant state. Finally, Ks was computed with equation 15.
Where Q = volume of water flowing through the sample, L = the length of soil sample, A = the cross-
section surface of sample, t = the length of time lapse and h = the constant column height of the water
above the sample.
Bulk Density and Porosity
Subsequent to the Ks measurement Db and porosity were measured following the method introduced by
Soil Survey Staff (2014). For Db the core samples dried in an oven for 24 h at 105 °C. After removing from
the oven and cooling for approximately 30 min the core samples were weighted. Equation 16 and equation
17 were used to calculated Db and porosity, respectively.
Db (g cm−3)=Wd
V (16)
Where 𝑊𝑑= weight of dry soil (g) and V = volume of the sampling ring (cm3).
Porosity (%) = 100 − (𝐷𝑏
Pd∗ 100) (17)
Where Pd = Particle Density; an average value of 2.65 g cm-3 was used.
K𝑠 (mm h−1) =Q ∗ L
A ∗ t ∗ h (15)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
82
Particle Size Distribution (PSD)
PSD Analysis was used to assess the size distribution of the individual soil particles in each disturbed soil
sample, in general terms, the percentage of silt, sand and clay particles. Using the Hydrometer Method as
published by the Soil Science Society of America and outlined in the Soil Survey and Laboratory Manual
(Soil Survey Staff, 2014), the PSD was assessed. The air-dried and grinded disturbed samples were passed
through a 2 mm sieve. 40 g of each sample was mixed in a glass beaker with 100 ml of distilled water and
100 ml of a 5 % sodium hexametaphosphate solution (HMP). The HMP solution is applied to attain
complete dispersal of the soil particles. The samples were left over night to soak and then stirred with an
industrial mixer for 5 min. In a sedimentation cylinder which was filled up to the 1 L mark, the samples
could equilibrate thermally. Parallel a sedimentation cylinder with a blank HMP solution was prepared for
hydrometer reading correction computation. In a manual up-and-down motion, the samples were stirred
again for 30 sec. Immediately, after stirring, time was recorded, temperature measured and a hydrometer
(ASTM 152H) inserted. Hydrometer and thermometer readings were taken at 30 sec, 60 sec, 3 min, 10 min,
30 min, 60 min, 90 min, 120 min and 24 h for the soil and blank solution. Percentages of sand silt and clay
were subsequently computed as suggested in the Soil Survey and Laboratory Manual (Soil Survey Staff,
2014).
Soil Organic Matter
For the determination of the SOM content of the disturbed surface samples the Loss-on-Ignition (LOI)
method as set out by Schulte and Hopkins (1996) was applied. LOI was chosen as it reprents an even more
effective and simpler way to determine SOM compared to other conventional methods such as the Walkley
and Black Carbon (WB-C) method which requieres additional chemicals and laboratory facilities
(Paramananthan et al., 2018). As pre-treatment the soil samples (5 g) where dried by 105 °C for 24 h in an
oven to remove all available water. In a desiccator the samples cooled down and were weighted before going
to the furnace. The furnace was pre-heated to have constant temperature of 400 °C. For 8 h the samples
were exposed to the furnace before cooling down to room temperature in a desiccator again. The percentage
of SOM was then determined by calculating the weight loss between oven-dry soil state and furnace ignition
state.
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
83
Annex 2.5. Physical and chemical properties of SoilGrids for each site
Site
Slope
Position Land Cover
Clay
(%)
Sand
(%)
Silt
(%)
Ks
(mm h-1)
Porosity
(%)
Bulk Density
(g cm-3)
OM
(%)
Soil Textural
Class
1 Backslope Orchard 24 41 34 36 59 1.5 5.8 Loam 2 Summit Cropland 26 40 33 32 59 1.5 6.2 Loam 3 Valley Cropland 29 39 32 32 59 1.5 6.2 Clay Loam 4 Valley Cropland 29 39 32 31 58 1.6 5.8 Clay Loam 5 Backslope Orchard 25 41 34 23 55 1.5 4.6 Loam 6 Backslope Cropland 22 43 35 57 65 1.5 8.3 Loam 7 Valley Cropland 25 41 34 41 58 1.4 5.3 Loam 8 Valley Mixed Forest 27 42 31 26 56 1.5 4.8 Loam 9 Summit Mixed Forest 24 43 33 41 61 1.3 7.1 Loam 10 Valley Orchard 30 38 32 28 57 1.4 5.7 Clay Loam 11 Valley Teak Plantation 26 41 33 22 56 1.5 5.0 Loam 12 Summit Mixed Forest 26 41 34 34 57 1.6 4.8 Loam 13 Backslope Mixed Forest 28 42 31 29 57 1.5 5.2 Clay Loam 14 Backslope Mixed Forest 28 42 30 36 58 1.5 5.2 Clay Loam 15 Valley Cropland 28 42 31 36 56 1.6 4.6 Clay Loam 16 Valley Mixed Forest 30 39 32 23 56 1.5 4.5 Clay Loam 17 Backslope Mixed Forest 28 42 30 23 56 1.5 4.5 Clay Loam 18 Backslope Mixed Forest 29 38 33 23 56 1.5 4.5 Clay Loam 19 Backslope Orchard 25 46 30 37 59 1.6 6.2 Loam 20 Backslope Orchard 24 43 33 36 60 1.6 6.7 Loam 21 Backslope Orchard 24 43 33 45 60 1.4 6.7 Loam 22 Valley Orchard 24 44 32 36 60 1.6 6.7 Loam 23 Backslope Orchard 26 42 33 36 61 1.6 7.1 Loam 24 Summit Orchard 26 42 32 41 61 1.4 7.2 Loam 25 Valley Orchard 26 42 33 36 61 1.6 7.1 Loam 26 Valley Orchard 27 41 32 34 59 1.5 6.5 Loam 27 Backslope Orchard 27 41 32 34 59 1.5 6.5 Loam 28 Summit Orchard 24 43 33 32 59 1.5 6.7 Loam 29 Valley Orchard 26 44 30 34 58 1.5 5.7 Loam 30 Valley Orchard 24 41 34 35 58 1.5 5.5 Loam 31 Backslope Orchard 25 40 34 27 55 1.5 4.5 Loam 32 Backslope Orchard 25 43 32 27 55 1.5 4.5 Loam 33 Valley Orchard 25 40 34 37 59 1.5 5.7 Loam 34 Valley Orchard 27 40 34 42 60 1.4 6.5 Loam 35 Backslope Orchard 25 42 32 42 60 1.4 6.5 Loam 36 Backslope Orchard 25 42 32 37 61 1.4 6.9 Loam 37 Valley Orchard 28 39 33 40 62 1.4 7.4 Clay Loam 38 Backslope Orchard 26 40 34 41 62 1.5 7.1 Loam 39 Summit Orchard 27 39 34 41 61 1.5 6.7 Loam 40 Backslope Orchard 23 44 34 41 61 1.5 6.7 Loam 41 Valley Orchard 23 44 34 41 61 1.5 6.7 Loam 42 Backslope Teak Plantation 30 37 33 22 56 1.5 4.6 Clay Loam 43 Backslope Orchard 25 41 34 28 56 1.6 5.0 Loam 44 Backslope Orchard 28 41 30 31 59 1.6 6.4 Clay Loam 45 Backslope Orchard 27 41 32 30 57 1.5 5.0 Loam 46 Backslope Orchard 25 43 31 35 60 1.4 6.5 Loam 47 Valley Orchard 26 41 33 39 57 1.4 4.8 Loam 48 Backslope Orchard 29 39 32 17 53 1.6 3.6 Clay Loam
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
84
Annex 3. Description of Landsat images used for yearly and seasonal composites
Median composite 2005/06
ID Date Satellite
LANDSAT/LT05/C01/T1_SR/LT05_130048_20050131 31/01/2005
Lan
dsa
t 5
LANDSAT/LT05/C01/T1_SR/LT05_130048_20050216 16/02/2005
LANDSAT/LT05/C01/T1_SR/LT05_130048_20050421 21/04/2005
LANDSAT/LT05/C01/T1_SR/LT05_130048_20060203 03/02/2006
LANDSAT/LT05/C01/T1_SR/LT05_130048_20060307 07/03/2006
Median composite dry cool season
Year ID Date Satellite
2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180119 19/01/2018
Lan
dsa
t 8
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180204 04/02/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180220 20/02/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181103 03/11/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181119 19/11/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181205 05/12/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181221 21/12/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20190207 07/02/2019
2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170116 16/01/2017
Lan
dsa
t 8
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170201 01/02/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170217 17/02/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20171202 02/12/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20171218 18/12/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180119 19/01/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180204 04/02/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180220 20/02/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181103 03/11/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181119 19/11/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181205 05/12/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181221 21/12/2018
Median composite rainy season
Year ID Date Satellite
2018 LANDSAT/LC08/C01/T1_SR/LC08_130048_20180916 16/09/2018
Lan
dsa
t
8
LANDSAT/LC08/C01/T1_SR/LC08_130048_20181018 18/10/2018
2017 LANDSAT/LC08/C01/T1_SR/LC08_130048_20170913 13/09/2017
Lan
dsa
t
8
LANDSAT/LC08/C01/T1_SR/LC08_130048_20171031 31/10/2017
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
85
Median composite dry hot season
Year ID Date Satellite
2018 LANDSAT/LC08/C01/T1_SR/LC08_130048_20180308 08/03/2018
Lan
dsa
t 8
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180409 09/04/2018
LANDSAT/LC08/C01/T1_SR/LC08_130048_20180425 25/04/2018
2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170305 05/03/2017
Lan
dsa
t 8 LANDSAT/LC08/C01/T1_SR/LC08_130048_20170321 21/03/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170406 06/04/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170422 22/04/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170508 08/05/2017
LANDSAT/LC08/C01/T1_SR/LC08_130048_20170625 25/06/2017
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
86
Annex 3.1. Google Earth Engine code for cloud removal by Saah et al. (2019) // Compute a cloud score and adds a band that represents the cloud mask. // This expects the input image to have the common band names: // ["red", "blue", etc], so it can work across sensors. exports.landsatCloudScore = function(ls8,cloudScoreThresh,cloudScorePctl,contractPixels,dilatePixels){ function getCloudScore (img) { // Compute several indicators of cloudiness and take the minimum of them. var score = ee.Image(1.0); // Clouds are reasonably bright in the blue band. score = score.min(rescale(img, 'img.blue', [0.1, 0.3])); // Clouds are reasonably bright in all visible bands. score = score.min(rescale(img, 'img.red + img.green + img.blue', [0.2, 0.8])); // Clouds are reasonably bright in all infrared bands. score = score.min( rescale(img, 'img.nir + img.swir1 + img.swir2', [0.3, 0.8])); // Clouds are reasonably cool in temperature. score = score.min(rescale(img,'img.temp', [300, 290])); // However, clouds are not snow. var ndsi = img.normalizedDifference(['green', 'swir1']); score = score.min(rescale(ndsi, 'img', [0.8, 0.6])); score = score.multiply(100).byte(); score = score.clamp(0,100); return img.addBands(score.rename(['cloudScore'])); } function maskScore(img){ var cloudMask = img.select(['cloudScore']).lt(cloudScoreThresh).focal_max(contractPixels).focal_min(dilatePixels).rename('cloudMask'); return img.updateMask(cloudMask).addBands(cloudMask); } ls8 = ls8.map(getCloudScore); // Find low cloud score pctl for each pixel to avoid comission errors var minCloudScore = ls8.select(['cloudScore']).reduce(ee.Reducer.percentile([cloudScorePctl])); ls8 = ls8.map(maskScore); return ls8 }; exports.QAMaskCloud = function(ls8){ // Functions for applying fmask to SR data with QA band var fmaskBitDict = {'cloud' : 32, 'shadow': 8,'snow':16}; function cFmask(img,fmaskClass){ var m = img.select('pixel_qa').bitwiseAnd(fmaskBitDict[fmaskClass]).neq(0); return img.updateMask(m.not()); } function cFmaskCloud(img){ return cFmask(img,'cloud'); } function cFmaskCloudShadow(img){ return cFmask(img,'shadow'); } ls8 = ls8.map(cFmaskCloud).map(cFmaskCloudShadow) return ls8 };
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
87
Annex 3.2. Google Earth Engine code for cloud shadow removal by Saah et al. (2019)
// Function for finding dark outliers in time series.
// Original concept written by Carson Stam and adapted by Ian Housman.
// Adds a band that is a mask of pixels that are dark, and dark outliers.
exports.shadowMask =
function(collection,studyArea,zScoreThresh,shadowSumThresh,contractPixels,dilatePixels) {
var shadowSumBands = ['nir','swir1'];
var allCollection = collection.filterBounds(studyArea).select(shadowSumBands);
// Get some pixel-wise stats for the time series
var irStdDev = allCollection.select(shadowSumBands).reduce(ee.Reducer.stdDev());
var irMean = allCollection.select(shadowSumBands).mean();
var maskDarkOutliers = function(img){
var zScore = img.select(shadowSumBands).subtract(irMean).divide(irStdDev);
var irSum = img.select(shadowSumBands).reduce(ee.Reducer.sum());
var TDOMMask =
zScore.lt(zScoreThresh).reduce(ee.Reducer.sum()).eq(2).and(irSum.lt(shadowSumThresh));
TDOMMask =
TDOMMask.focal_min(contractPixels).focal_max(dilatePixels).rename('TDOMMask');
return img.updateMask(TDOMMask.not()).addBands(TDOMMask);
};
// Mask out dark dark outliers
collection = collection.map(maskDarkOutliers)
return collection;
};
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
88
Annex 3.3. Google Earth Engine code for BRDF correction by Saah et al. (2019)
var PI = ee.Number(3.14159265359);
var MAX_SATELLITE_ZENITH = 7.5;
var MAX_DISTANCE = 1000000;
var UPPER_LEFT = 0;
var LOWER_LEFT = 1;
var LOWER_RIGHT = 2;
var UPPER_RIGHT = 3;
exports.brdfS2 = function(collection) {
collection = collection.map(applyBRDF);
return collection;
function applyBRDF(image){
var date = image.date();
var footprint = ee.List(image.geometry().bounds().bounds().coordinates().get(0));
var angles = getsunAngles(date, footprint);
var sunAz = angles[0];
var sunZen = angles[1];
var viewAz = azimuth(footprint);
var viewZen = zenith(footprint);
var kval = _kvol(sunAz, sunZen, viewAz, viewZen);
var kvol = kval[0];
var kvol0 = kval[1];
var result = _apply(image, kvol.multiply(PI), kvol0.multiply(PI));
return result;}
/* Get sunAngles from the map given the data. */
function getsunAngles(date, footprint){
var jdp = date.getFraction('year');
var seconds_in_hour = 3600;
var hourGMT = ee.Number(date.getRelative('second', 'day')).divide(seconds_in_hour);
var latRad = ee.Image.pixelLonLat().select('latitude').multiply(PI.divide(180));
var longDeg = ee.Image.pixelLonLat().select('longitude');
// Julian day proportion in radians
var jdpr = jdp.multiply(PI).multiply(2);
var a = ee.List([0.000075, 0.001868, 0.032077, 0.014615, 0.040849]);
var meanSolarTime = longDeg.divide(15.0).add(ee.Number(hourGMT));
var localSolarDiff1 = value(a, 0)
.add(value(a, 1).multiply(jdpr.cos()))
.subtract(value(a, 2).multiply(jdpr.sin()))
.subtract(value(a, 3).multiply(jdpr.multiply(2).cos()))
.subtract(value(a, 4).multiply(jdpr.multiply(2).sin()));
var localSolarDiff2 = localSolarDiff1.multiply(12 * 60);
var localSolarDiff = localSolarDiff2.divide(PI);
var trueSolarTime = meanSolarTime
.add(localSolarDiff.divide(60)) .subtract(12.0);
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
89
// Hour as an angle;
var ah = trueSolarTime.multiply(ee.Number(MAX_SATELLITE_ZENITH *
2).multiply(PI.divide(180))) ;
var b = ee.List([0.006918, 0.399912, 0.070257, 0.006758, 0.000907, 0.002697, 0.001480]);
var delta = value(b, 0)
.subtract(value(b, 1).multiply(jdpr.cos()))
.add(value(b, 2).multiply(jdpr.sin()))
.subtract(value(b, 3).multiply(jdpr.multiply(2).cos()))
.add(value(b, 4).multiply(jdpr.multiply(2).sin()))
.subtract(value(b, 5).multiply(jdpr.multiply(3).cos()))
.add(value(b, 6).multiply(jdpr.multiply(3).sin()));
var cosSunZen = latRad.sin().multiply(delta.sin())
.add(latRad.cos().multiply(ah.cos()).multiply(delta.cos()));
var sunZen = cosSunZen.acos();
// sun azimuth from south, turning west
var sinSunAzSW = ah.sin().multiply(delta.cos()).divide(sunZen.sin());
sinSunAzSW = sinSunAzSW.clamp(-1.0, 1.0);
var cosSunAzSW = (latRad.cos().multiply(-1).multiply(delta.sin())
.add(latRad.sin().multiply(delta.cos()).multiply(ah.cos())))
.divide(sunZen.sin());
var sunAzSW = sinSunAzSW.asin();
sunAzSW = where(cosSunAzSW.lte(0), sunAzSW.multiply(-1).add(PI), sunAzSW);
sunAzSW = where(cosSunAzSW.gt(0).and(sinSunAzSW.lte(0)), sunAzSW.add(PI.multiply(2)),
sunAzSW);
var sunAz = sunAzSW.add(PI);
// # Keep within [0, 2pi] range
sunAz = where(sunAz.gt(PI.multiply(2)), sunAz.subtract(PI.multiply(2)), sunAz);
var footprint_polygon = ee.Geometry.Polygon(footprint);
sunAz = sunAz.clip(footprint_polygon);
sunAz = sunAz.rename(['sunAz']);
sunZen = sunZen.clip(footprint_polygon).rename(['sunZen']);
return [sunAz, sunZen];
}
/* Get azimuth. */
function azimuth(footprint){
function x(point){return ee.Number(ee.List(point).get(0))}
function y(point){return ee.Number(ee.List(point).get(1))}
var upperCenter = line_from_coords(footprint, UPPER_LEFT,
UPPER_RIGHT).centroid().coordinates();
var lowerCenter = line_from_coords(footprint, LOWER_LEFT,
LOWER_RIGHT).centroid().coordinates();
var slope =
((y(lowerCenter)).subtract(y(upperCenter))).divide((x(lowerCenter)).subtract(x(upperCenter)));
var slopePerp = ee.Number(-1).divide(slope);
var azimuthLeft = ee.Image(PI.divide(2).subtract((slopePerp).atan()));
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
90
return azimuthLeft.rename(['viewAz']);
}
/* Get zenith. */
function zenith(footprint){
var leftLine = line_from_coords(footprint, UPPER_LEFT, LOWER_LEFT);
var rightLine = line_from_coords(footprint, UPPER_RIGHT, LOWER_RIGHT);
var leftDistance = ee.FeatureCollection(leftLine).distance(MAX_DISTANCE);
var rightDistance = ee.FeatureCollection(rightLine).distance(MAX_DISTANCE);
var viewZenith = rightDistance.multiply(ee.Number(MAX_SATELLITE_ZENITH * 2))
.divide(rightDistance.add(leftDistance))
.subtract(ee.Number(MAX_SATELLITE_ZENITH))
.clip(ee.Geometry.Polygon(footprint))
.rename(['viewZen']);
return viewZenith.multiply(PI.divide(180));
}
/* apply function to all bands
function _apply(image, kvol, kvol0){
var f_iso = 0;
var f_geo = 0;
var f_vol = 0;
var blue = _correct_band(image, 'blue', kvol, kvol0, f_iso=0.0774, f_geo=0.0079, f_vol=0.0372);
var green = _correct_band(image, 'green', kvol, kvol0, f_iso=0.1306, f_geo=0.0178, f_vol=0.0580);
var red = _correct_band(image, 'red', kvol, kvol0, f_iso=0.1690, f_geo=0.0227, f_vol=0.0574);
var re1 = _correct_band(image, 're1', kvol, kvol0, f_iso=0.2085, f_geo=0.0256, f_vol=0.0845);
var re2 = _correct_band(image, 're2', kvol, kvol0, f_iso=0.2316, f_geo=0.0273, f_vol=0.1003);
var re3 = _correct_band(image, 're3', kvol, kvol0, f_iso=0.2599, f_geo=0.0294, f_vol=0.1197);
var nir = _correct_band(image, 'nir', kvol, kvol0, f_iso=0.3093, f_geo=0.0330, f_vol=0.1535);
var re4 = _correct_band(image, 're4', kvol, kvol0, f_iso=0.2907, f_geo=0.0410, f_vol=0.1611);
var swir1 = _correct_band(image, 'swir1', kvol, kvol0, f_iso=0.3430, f_geo=0.0453, f_vol=0.1154);
var swir2 = _correct_band(image, 'swir2', kvol, kvol0, f_iso=0.2658, f_geo=0.0387, f_vol=0.0639);
var temp = image.select('temp');
return image.select([]).addBands([blue, green, red, nir,re1,re2,re3,nir,re4,swir1,swir2,temp]);
}
/* correct band function
function _correct_band(image, band_name, kvol, kvol0, f_iso, f_geo, f_vol){
//"""fiso + fvol * kvol + fgeo * kgeo"""
var iso = ee.Image(f_iso);
var geo = ee.Image(f_geo);
var vol = ee.Image(f_vol);
var pred = vol.multiply(kvol).add(geo.multiply(kvol)).add(iso).rename(['pred']);
var pred0 = vol.multiply(kvol0).add(geo.multiply(kvol0)).add(iso).rename(['pred0']);
var cfac = pred0.divide(pred).rename(['cfac']);
var corr = image.select(band_name).multiply(cfac).rename([band_name]);
return corr;
}
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
91
/* calculate kvol and kvol0 */
function _kvol(sunAz, sunZen, viewAz, viewZen){
var relative_azimuth = sunAz.subtract(viewAz).rename(['relAz']);
var pa1 = viewZen.cos().multiply(sunZen.cos());
var pa2 = viewZen.sin().multiply(sunZen.sin()).multiply(relative_azimuth.cos());
var phase_angle1 = pa1.add(pa2);
var phase_angle = phase_angle1.acos();
var p1 = ee.Image(PI.divide(2)).subtract(phase_angle);
var p2 = p1.multiply(phase_angle1);
var p3 = p2.add(phase_angle.sin());
var p4 = sunZen.cos().add(viewZen.cos());
var p5 = ee.Image(PI.divide(4));
var kvol = p3.divide(p4).subtract(p5).rename(['kvol']);
var viewZen0 = ee.Image(0);
var pa10 = viewZen0.cos().multiply(sunZen.cos());
var pa20 = viewZen0.sin().multiply(sunZen.sin()).multiply(relative_azimuth.cos());
var phase_angle10 = pa10.add(pa20);
var phase_angle0 = phase_angle10.acos();
var p10 = ee.Image(PI.divide(2)).subtract(phase_angle0);
var p20 = p10.multiply(phase_angle10);
var p30 = p20.add(phase_angle0.sin());
var p40 = sunZen.cos().add(viewZen0.cos());
var p50 = ee.Image(PI.divide(4));
var kvol0 = p30.divide(p40).subtract(p50).rename(['kvol0']);
return [kvol, kvol0]}
/* helper function */
function line_from_coords(coordinates, fromIndex, toIndex){
return ee.Geometry.LineString(ee.List([
coordinates.get(fromIndex),
coordinates.get(toIndex)]));
}
function where(condition, trueValue, falseValue){
var trueMasked = trueValue.mask(condition);
var falseMasked = falseValue.mask(invertMask(condition));
return trueMasked.unmask(falseMasked);
}
function invertMask(mask){
return mask.multiply(-1).add(1);
}
function value(list,index){
return ee.Number(list.get(index));
}
};
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
92
Annex 3.4. Google Earth Engine code for topographic correction by Saah et al. (2019)
ar scale = 300;
var toaOrSR = 'SR';
// get terrain layers
var dem = ee.Image("USGS/SRTMGL1_003")
var degree2radian = 0.01745;
exports.terrainCorrection = function(collection) {
function getTopo(img){
/* function to filter for areas with terrain and areas without */
dem = dem.unmask(0);
var geom = ee.Geometry(img.get('system:footprint')).bounds();
print(geom)
var slp_rad = ee.Terrain.slope(dem).clip(geom);
var slope = slp_rad.reduceRegion({reducer: ee.Reducer.percentile([80]),geometry: geom,scale: 100});
return img.set('slope',slope.get('slope'));
}
function pixelArea(img){
/* check if there is data in the image */
var geom = ee.Geometry(img.get('system:footprint')).bounds();
var area = img.select(['red']).gt(0).reduceRegion({reducer:ee.Reducer.sum(),geometry: geom,scale:100});
return img.set("pixelArea",area.get("red"))}
collection = collection.map(pixelArea);
collection = collection.filter(ee.Filter.gt("pixelArea",100));
//collection = collection.map(getTopo);
var correction = collection.filter(ee.Filter.gte("slope",10));
var notcorrection = collection.filter(ee.Filter.lt("slope",10));
collection = collection.map(illuminationCondition);
collection = collection.map(illuminationCorrection);
return(collection);
// Extract image metadata about solar position
var SZ_rad =
ee.Image.constant(ee.Number(img.get('SOLAR_ZENITH_ANGLE'))).multiply(3.14159265359).divide(18
0).clip(img.geometry().buffer(10000));
var SA_rad =
ee.Image.constant(ee.Number(img.get('SOLAR_AZIMUTH_ANGLE')).multiply(3.14159265359).divide(1
80)).clip(img.geometry().buffer(10000));
// Creat terrain layers
var slp = ee.Terrain.slope(dem).clip(img.geometry().buffer(10000));
var slp_rad =
ee.Terrain.slope(dem).multiply(3.14159265359).divide(180).clip(img.geometry().buffer(10000));
var asp_rad =
ee.Terrain.aspect(dem).multiply(3.14159265359).divide(180).clip(img.geometry().buffer(10000));
// Calculate the Illumination Condition (IC) slope part of the illumination condition
var cosZ = SZ_rad.cos();
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
93
var cosS = slp_rad.cos();
var slope_illumination = cosS.expression("cosZ * cosS",
{'cosZ': cosZ,
'cosS': cosS.select('slope')});
// aspect part of the illumination condition
var sinZ = SZ_rad.sin();
var sinS = slp_rad.sin();
var cosAziDiff = (SA_rad.subtract(asp_rad)).cos();
var aspect_illumination = sinZ.expression("sinZ * sinS * cosAziDiff",
{'sinZ': sinZ,
'sinS': sinS,
'cosAziDiff': cosAziDiff});
// full illumination condition (IC)
var ic = slope_illumination.add(aspect_illumination);
// Add IC to original image
var img_plus_ic =
ee.Image(img.addBands(ic.rename('IC')).addBands(cosZ.rename('cosZ')).addBands(cosS.rename('cosS')).ad
dBands(slp.rename('slope')));
return img_plus_ic;
}
// Function to apply the Sun-Canopy-Sensor + C (SCSc) correction method to each image
function illuminationCorrection(img){
var props = img.toDictionary();
var st = img.get('system:time_start');
var img_plus_ic = img;
var mask1 = img_plus_ic.select('nir').gt(-0.1);
var mask2 = img_plus_ic.select('slope').gte(5)
.and(img_plus_ic.select('IC').gte(0))
.and(img_plus_ic.select('nir').gt(-0.1));
var img_plus_ic_mask2 = ee.Image(img_plus_ic.updateMask(mask2));
// Specify Bands to topographically correct
var bandList = ['blue','green','red','nir','swir1','swir2'];
var compositeBands = img.bandNames();
var nonCorrectBands = img.select(compositeBands.removeAll(bandList));
var geom = ee.Geometry(img.get('system:footprint')).bounds().buffer(10000);
function apply_SCSccorr(band){
var method = 'SCSc';
var out = img_plus_ic_mask2.select('IC', band).reduceRegion({
reducer: ee.Reducer.linearFit(),
// Compute coefficients:
a(slope), b(offset), c(b/a) geometry: ee.Geometry(img.geometry().buffer(-5000)),
scale: 300,
maxPixels: 1000000000
});
if (out === null || out === undefined ){
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
94
return img_plus_ic_mask2.select(band);
}
else{
var out_a = ee.Number(out.get('scale'));
var out_b = ee.Number(out.get('offset'));
var out_c = out_b.divide(out_a);
// Apply the SCSc correction
var SCSc_output = img_plus_ic_mask2.expression(
"((image * (cosB * cosZ + cvalue)) / (ic + cvalue))", {
'image': img_plus_ic_mask2.select(band),
'ic': img_plus_ic_mask2.select('IC'),
'cosB': img_plus_ic_mask2.select('cosS'),
'cosZ': img_plus_ic_mask2.select('cosZ'),
'cvalue': out_c
});
return SCSc_output;
}
}
var img_SCSccorr = ee.Image(bandList.map(apply_SCSccorr)).addBands(img_plus_ic.select('IC'));
var bandList_IC = ee.List([bandList, 'IC']).flatten();
img_SCSccorr = img_SCSccorr.unmask(img_plus_ic.select(bandList_IC)).select(bandList);
return img_SCSccorr.addBands(nonCorrectBands)
.setMulti(props)
.set('system:time_start',st);
}
}
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
95
Annex 3.5. Google Earth Engine code to generate covariate layer (seasonal) adapted from Saah et al. (2019) var elevation = ee.Image("USGS/SRTMGL1_003"); var jrcImage = ee.Image("JRC/GSW1_0/GlobalSurfaceWater"); var ndCovariatesList = [ ['blue', 'green'], ['blue', 'red'], ['blue', 'nir'], ['blue', 'swir1'], ['blue', 'swir2'], ['green', 'red'], ['green', 'nir'], ['green', 'swir1'], ['green', 'swir2'], ['red', 'swir1'], ['red', 'swir2'], ['nir', 'red'], ['nir', 'swir1'], ['nir', 'swir2'], ['swir1', 'swir2'] ]; var rCovariatesList = [ ['swir1', 'nir'], ['red', 'swir1'] ]; var ComputeNDCovariatesList = function (season) { var list = []; for (var index in ndCovariatesList) { var list_ = [season + '_' + ndCovariatesList[index][0], season + '_' + ndCovariatesList[index][1]]; list.push(list_); } return list; }; var addNDCovariates = function (season, image){ var list = ComputeNDCovariatesList(season); for (var index in list) { image = image.addBands(image.normalizedDifference(list[index]).rename(season + '_ND_'+ ndCovariatesList[index][0] + '_' + ndCovariatesList[index][1])); } return image; }; var ComputeRCovariatesList = function (season) { var list = []; for (var index in rCovariatesList) { var list_ = [season + '_' + rCovariatesList[index][0], season + '_' + rCovariatesList[index][1]]; list.push(list_); } return list; }; var addRCovariates = function (season, image) { var list = ComputeRCovariatesList(season); for (var index in list) { image = image.addBands(image.select(list[index][0]).divide(image.select(list[index][1])) .rename(season + '_R_' + rCovariatesList[index][0] + '_' + rCovariatesList[index][1])); }
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
96
return image; }; //Enhanced Vegetation Index (EVI)// var addEVI = function (season, image) { var evi = image.expression('2.5 * ((NIR - RED) / (NIR + 6 * RED - 7.5 * BLUE + 1))', { 'NIR' : image.select(season + '_nir'), 'RED' : image.select(season + '_red'), 'BLUE': image.select(season + '_blue') }).float(); return image.addBands(evi.rename(season + '_EVI')); }; //Enhanced Built-Up and Bareness Index (EBBI)// var addEBBI = function (season, image) { print(image); var ebbi= image.expression("(swir-nir) / 10*sqrt(swir+temp)",{ 'swir' : image.select(season + '_swir1'), 'nir' : image.select(season + '_nir'), 'temp' : image.select(season + '_temp'), }); return image.addBands(ebbi.rename(season + '_EBBI')); }; //Normalized Burn Index (NBR)// var addNBR = function (season, image) { var nbr = image.normalizedDifference([season + '_nir', season + '_swir2']).float(); return image.addBands(nbr.rename(season + '_NBR')); }; //Normalized Difference Vegetation Index// var addNDVI = function(season, image) { var NDVI =image.normalizedDifference([season + '_red', season + '_nir']).float(); return image.addBands(NDVI.rename(season + '_NDVI')); }; //Soil Adjusted Vegetation Index (SAVI)// var addSAVI = function (season, image) { // Add Soil Adjust Vegetation Index (SAVI) // using L = 0.5; var savi = image.expression('(NIR - RED) * (1 + 0.5)/(NIR + RED + 0.5)', { 'NIR': image.select(season + '_nir'), 'RED': image.select(season + '_red') }).float(); return image.addBands(savi.rename(season + '_SAVI')); }; // Index-based Built-up Index// var addIBI = function (season, image) { var ibiA = image.expression('2 * SWIR1 / (SWIR1 + NIR)', { 'SWIR1': image.select(season + '_swir1'), 'NIR' : image.select(season + '_nir') }).rename(['IBI_A']); var ibiB = image.expression('(NIR / (NIR + RED)) + (GREEN / (GREEN + SWIR1))', { 'NIR' : image.select(season + '_nir'), 'RED' : image.select(season + '_red'), 'GREEN': image.select(season + '_green'), 'SWIR1': image.select(season + '_swir1') }).rename(['IBI_B']); var ibiAB = ibiA.addBands(ibiB); var ibi = ibiAB.normalizedDifference(['IBI_A', 'IBI_B']);
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
97
return image.addBands(ibi.rename([season + '_IBI'])); }; // Function to compute the Tasseled Cap transformation// var getTassledCapComponents = function (season, image) { var coefficients = ee.Array([ [0.3037, 0.2793, 0.4743, 0.5585, 0.5082, 0.1863], [-0.2848, -0.2435, -0.5436, 0.7243, 0.0840, -0.1800], [0.1509, 0.1973, 0.3279, 0.3406, -0.7112, -0.4572], [-0.8242, 0.0849, 0.4392, -0.0580, 0.2012, -0.2768], [-0.3280, 0.0549, 0.1075, 0.1855, -0.4357, 0.8085], [0.1084, -0.9022, 0.4120, 0.0573, -0.0251, 0.0238] ]); var bands = ee.List([season + '_blue', season + '_green', season + '_red', season + '_nir', season + '_swir1', season + '_swir2']); var arrayImage1D = image.select(bands).toArray(); var arrayImage2D = arrayImage1D.toArray(1); var componentsImage = ee.Image(coefficients).matrixMultiply(arrayImage2D).arrayProject([0]) .arrayFlatten([[season + '_brightness', season + '_greenness', season + '_wetness', season + '_fourth', season + '_fifth', season + '_sixth']]).float(); return image.addBands(componentsImage); }; var getTassledCapAngleAndDistance = function (season, image) { var brightness = image.select(season + '_brightness'); var greenness = image.select(season + '_greenness'); var wetness = image.select(season + '_wetness'); var tcAngleBG = brightness.atan2(greenness).divide(Math.PI).rename([season + '_tcAngleBG']); var tcAngleGW = greenness.atan2(wetness).divide(Math.PI).rename([season + '_tcAngleGW']); var tcAngleBW = brightness.atan2(wetness).divide(Math.PI).rename([season + '_tcAngleBW']); var tcDistanceBG = brightness.hypot(greenness).rename([season + '_tcDistanceBG']); var tcDistanceGW = greenness.hypot(wetness).rename([season + '_tcDistanceGW']); var tcDistanceBW = brightness.hypot(wetness).rename([season + '_tcDistanceBW']); image = image.addBands(tcAngleBG).addBands(tcAngleGW).addBands(tcAngleBW).addBands(tcDistanceBG).addBands(tcDistanceGW).addBands(tcDistanceBW); return image; }; var computeTassledCap = function (season, image) { image = getTassledCapComponents(season, image); image = getTassledCapAngleAndDistance(season, image); return image; }; var addTopography = function (image) { // Calculate slope, aspect and hillshade var topo = ee.Algorithms.Terrain(elevation); // From aspect (a), calculate eastness (sin a), northness (cos a) var deg2rad = ee.Number(Math.PI).divide(180); var aspect = topo.select(['aspect']); var aspect_rad = aspect.multiply(deg2rad); var eastness = aspect_rad.sin().rename(['eastness']).float(); var northness = aspect_rad.cos().rename(['northness']).float(); topo = topo.select(['elevation','slope','aspect']).addBands(eastness).addBands(northness); image = image.addBands(topo); return image; }; //JRC dataset//
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
98
var addJRCDataset = function (image) { // Update the mask. jrcImage = jrcImage.unmask(0); image = image.addBands(jrcImage.select(['occurrence']).rename(['occurrence'])); image = image.addBands(jrcImage.select(['change_abs']).rename(['change_abs'])); image = image.addBands(jrcImage.select(['change_norm']).rename(['change_norm'])); image = image.addBands(jrcImage.select(['seasonality']).rename(['seasonality'])); image = image.addBands(jrcImage.select(['transition']).rename(['transition'])); image = image.addBands(jrcImage.select(['max_extent']).rename(['max_extent'])); return image; }; //Add covariates to season images// var addCovariates = function (season, image) { image = addNDCovariates(season, image); image = addRCovariates(season, image); image = addNDVI(season, image); image = addEVI(season, image); image = addSAVI(season, image); image = addIBI(season, image); image = addEBBI(season, image); image = addNBR(season, image); image = computeTassledCap(season, image); return image; }; var addJRCAndTopo = function (image) { image = addTopography(image); image = addJRCDataset(image); return image; }; exports.addCovariates = addCovariates;
exports.addJRCAndTopo = addJRCAndTopo;
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
99
Annex 3.6. Google Earth Engine code to generate covariate layer (yearly) adapted from Saah et al. (2019) ar elevation = ee.Image("USGS/SRTMGL1_003"); var jrcImage = ee.Image("JRC/GSW1_0/GlobalSurfaceWater"); var ndCovariatesList = [ ['blue', 'green'], ['blue', 'red'], ['blue', 'nir'], ['blue', 'swir1'], ['blue', 'swir2'], ['green', 'red'], ['green', 'nir'], ['green', 'swir1'], ['green', 'swir2'], ['red', 'swir1'], ['red', 'swir2'], ['nir', 'red'], ['nir', 'swir1'], ['nir', 'swir2'], ['swir1', 'swir2'] ]; var rCovariatesList = [ ['swir1', 'nir'], ['red', 'swir1'] ]; var ComputeNDCovariatesList = function () { var list = []; for (var index in ndCovariatesList) { var list_ = [ndCovariatesList[index][0], ndCovariatesList[index][1]]; list.push(list_); } return list; }; var addNDCovariates = function (image){ var list = ComputeNDCovariatesList(); print(list); for (var index in list) { image = image.addBands(image.normalizedDifference(list[index]).rename('ND_'+ ndCovariatesList[index][0] + '_' + ndCovariatesList[index][1])); } return image; }; var ComputeRCovariatesList = function () { var list = []; for (var index in rCovariatesList) { var list_ = [rCovariatesList[index][0], rCovariatesList[index][1]]; list.push(list_); } return list; }; var addRCovariates = function (image) { var list = ComputeRCovariatesList(); for (var index in list) { image = image.addBands(image.select(list[index][0]).divide(image.select(list[index][1]))
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
100
.rename('_R_' + rCovariatesList[index][0] + '_' + rCovariatesList[index][1])); } return image; }; var addEVI = function (image) { var evi = image.expression('2.5 * ((NIR - RED) / (NIR + 6 * RED - 7.5 * BLUE + 1))', { 'NIR' : image.select('nir'), 'RED' : image.select('red'), 'BLUE': image.select('blue') }).float(); return image.addBands(evi.rename('EVI')); } var addEBBI = function (image) { print(image); var ebbi= image.expression("(swir-nir) / 10*sqrt(swir+temp)", {'swir' : image.select("swir1"), 'nir' : image.select("nir"), 'temp' : image.select("temp"), }); return image.addBands(ebbi.rename('EBBI')); }; var addNBR = function (image) { var nbr = image.normalizedDifference(['nir','swir2']).float(); return image.addBands(nbr.rename('NBR')); }; var addNDVI = function(image) { var NDVI =image.normalizedDifference(['red','nir']).float(); return image.addBands(NDVI.rename('NDVI')); }; var addSAVI = function (image) { // Add Soil Adjust Vegetation Index (SAVI) // using L = 0.5; var savi = image.expression('(NIR - RED) * (1 + 0.5)/(NIR + RED + 0.5)', { 'NIR': image.select('nir'), 'RED': image.select('red') }).float(); return image.addBands(savi.rename('SAVI')); }; var addIBI = function (image) { // Add Index-Based Built-Up Index (IBI) var ibiA = image.expression('2 * SWIR1 / (SWIR1 + NIR)', { 'SWIR1': image.select('swir1'), 'NIR' : image.select('nir') }).rename(['IBI_A']); var ibiB = image.expression('(NIR / (NIR + RED)) + (GREEN / (GREEN + SWIR1))', { 'NIR' : image.select('nir'), 'RED' : image.select('red'), 'GREEN': image.select('green'), 'SWIR1': image.select('swir1') }).rename(['IBI_B']); var ibiAB = ibiA.addBands(ibiB); var ibi = ibiAB.normalizedDifference(['IBI_A', 'IBI_B']); return image.addBands(ibi.rename(['IBI'])); };
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
101
// Function to compute the Tasseled Cap transformation and return an image var getTassledCapComponents = function (image) { var coefficients = ee.Array([ [0.3037, 0.2793, 0.4743, 0.5585, 0.5082, 0.1863], [-0.2848, -0.2435, -0.5436, 0.7243, 0.0840, -0.1800], [0.1509, 0.1973, 0.3279, 0.3406, -0.7112, -0.4572], [-0.8242, 0.0849, 0.4392, -0.0580, 0.2012, -0.2768], [-0.3280, 0.0549, 0.1075, 0.1855, -0.4357, 0.8085], [0.1084, -0.9022, 0.4120, 0.0573, -0.0251, 0.0238] ]); var bands = ee.List(['blue', 'green', 'red', 'nir', 'swir1', 'swir2']); // Make an Array Image, with a 1-D Array per pixel. var arrayImage1D = image.select(bands).toArray(); // Make an Array Image with a 2-D Array per pixel, 6 x 1 var arrayImage2D = arrayImage1D.toArray(1); var componentsImage = ee.Image(coefficients).matrixMultiply(arrayImage2D).arrayProject([0]) .arrayFlatten([['brightness', 'greenness', 'wetness', 'fourth', 'fifth', 'sixth']]).float(); // Get a multi-band image with TC-named bands return image.addBands(componentsImage); }; // Function to add Tasseled Cap angles and distances to an image. Assumes image has bands: 'brightness', 'greenness', and 'wetness'. var getTassledCapAngleAndDistance = function (image) { var brightness = image.select('brightness'); var greenness = image.select('greenness'); var wetness = image.select('wetness'); // Calculate tassled cap angles and distances var tcAngleBG = brightness.atan2(greenness).divide(Math.PI).rename(['tcAngleBG']); var tcAngleGW = greenness.atan2(wetness).divide(Math.PI).rename(['tcAngleGW']); var tcAngleBW = brightness.atan2(wetness).divide(Math.PI).rename(['tcAngleBW']); var tcDistanceBG = brightness.hypot(greenness).rename(['tcDistanceBG']); var tcDistanceGW = greenness.hypot(wetness).rename(['tcDistanceGW']); var tcDistanceBW = brightness.hypot(wetness).rename(['tcDistanceBW']); image = image.addBands(tcAngleBG).addBands(tcAngleGW).addBands(tcAngleBW).addBands(tcDistanceBG).addBands(tcDistanceGW).addBands(tcDistanceBW); return image; }; var computeTassledCap = function (image) { image = getTassledCapComponents(image); image = getTassledCapAngleAndDistance(image); return image; }; var addTopography = function (image) { // Calculate slope, aspect and hillshade var topo = ee.Algorithms.Terrain(elevation); // From aspect (a), calculate eastness (sin a), northness (cos a) var deg2rad = ee.Number(Math.PI).divide(180); var aspect = topo.select(['aspect']); var aspect_rad = aspect.multiply(deg2rad); var eastness = aspect_rad.sin().rename(['eastness']).float(); var northness = aspect_rad.cos().rename(['northness']).float(); // Add topography bands to image topo = topo.select(['elevation','slope','aspect']).addBands(eastness).addBands(northness);
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
102
image = image.addBands(topo); return image; }; var addJRCDataset = function (image) { // Update the mask. jrcImage = jrcImage.unmask(0); image = image.addBands(jrcImage.select(['occurrence']).rename(['occurrence'])); image = image.addBands(jrcImage.select(['change_abs']).rename(['change_abs'])); image = image.addBands(jrcImage.select(['change_norm']).rename(['change_norm'])); image = image.addBands(jrcImage.select(['seasonality']).rename(['seasonality'])); image = image.addBands(jrcImage.select(['transition']).rename(['transition'])); image = image.addBands(jrcImage.select(['max_extent']).rename(['max_extent'])); return image; }; exports.addCovariates = function (image) { image = addNDCovariates(image); image = addNDVI(image); image = addEVI(image); image = addNBR(image); image = addSAVI(image); image = addIBI(image); image = addEBBI(image); image = computeTassledCap(image); return image; }; var addJRCAndTopo = function (image) { image = addTopography(image); image = addJRCDataset(image); return image; }; exports.addJRCAndTopo = addJRCAndTopo;
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
103
Annex 3.7. Selected covariates layers for the classifications of 2005 and 2018
Covariate layers 2005 (single composite) Covariate layers 2018 (seasonal composites)
NBR dryCool_ND_green_swir1 ND_blue_green dryCool_ND_red_swir2 ND_blue_nir dryCool_brightness ND_blue_swir1 dryCool_cosS ND_green_nir dryCool_green ND_green_swir1 dryCool_nir ND_green_swir2 dryCool_red ND_nir_swir2 dryCool_sixth ND_red_swir1 dryCool_slope ND_swir1_swir2 dryCool_tcDistanceBW brightness dryHot_ND_blue_swir1 fifth dryHot_ND_green_swir1 fourth dryHot_ND_red_swir1 greenness dryHot_ND_red_swir2"," nir dryHot_brightness swir1 dryHot_nir tcAngleGW dryHot_sixth tcDistanceBG dryHot_swir1 tcDistanceGW dryHot_tcDistanceBG NBR dryHot_tcDistanceBW ND_blue_green dryHot_tcDistanceGW ND_blue_nir dryHot_wetness ND_blue_swir1 elevation rainy_EVI rainy_ND_blue_green rainy_ND_blue_nir rainy_ND_blue_red rainy_ND_blue_swir1 rainy_ND_green_nir rainy_ND_green_red rainy_ND_green_swir1 rainy_ND_green_swir2 rainy_ND_nir_red rainy_ND_nir_swir2 rainy_ND_red_swir1 rainy_ND_red_swir2 rainy_ND_swir1_swir2 rainy_R_red_swir1 rainy_SAVI rainy_cosS rainy_fifth rainy_fourth rainy_greenness rainy_red rainy_sixth rainy_swir1 rainy_tcAngleBG rainy_temp
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
104
Annex 4. IDF curves for Uttaradit province (Rittima et al., 2013)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
105
Annex 5. Stream dimension measurements, locations and average values for validation
Number X-UTM Y-UTM Depth Width
4.1 618205 1970317 2.8 14.76
4.2 619022 1968612 4.06 17.9
4.3 613011 1969383 0.6 2.2
5.1 617007 1965044 1.66 4.4
5.2 615851 1966393 1.67 5
5.3 617997 1969297 4.3 17.9
5.4 618126 1969100 3.8 27.2
6.1 617562 1965041 3.17 13.88
6.2 618275 1967013 3.15 16.85
Stream Order Depth average (m) Width average (m) Shape
4th 2.5 11.6 U
5th 2.9 13.6 U
6th 3.2 15.4 U
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
106
Annex 5.1. Channel maps Ban Da Na Kham a) and Laplae b) watershed
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
107
Annex 6. PCRaster script for soil depth map generation based on Kuriakose et al. (2009) #! --lddout --unittrue binding DEM = dem.map; soildepth = soildepth20m.map; LDD = ldd.map; soild = soild.map; initial mask=DEM/DEM; aspect = scalar(aspect(DEM))*mask; grad = max(0.001,slope(DEM))*100*mask; curv = profcurv(DEM); # LDD = lddcreate(DEM,1e20,1e20,1e20,1e20); stream = nominal(accuflux(LDD,celllength()) gt 2000); # used or accumulation of material, so choose larger rivers, not incising rivers Distance = spread(stream,0,1)*mask; wetness = ln(accuflux(LDD,cellarea())/(grad))*mask; demmax = mapmaximum(DEM); distmax = mapmaximum(Distance); report soild = cover( -0.1*DEM/demmax # lower altitudes give deeper soils -0.06*grad # steeper slopes giver undeep soils +0.01*wetness # higher wetness accumulates material, deeper soils -0*aspect # no aspect effect -0*curv # profile curv is - when convex, + when concave. Convex has deeper soils -0.2*Distance/distmax # perpendicular distance to river, closer gives deeper soils ,0)*mask; maxd = mapminimum(soild); soildb = 20*(soild-maxd)**1.1; # m to mm for lisem, higher power emphasizes deep, updeep report soildepth = windowaverage(soildb,5*celllength()); # smooth asp = cover(scalar( aspect(DEM)),0); # sine gradient (-), make sure slope > 0.001 shade = cos(15)*sin(grad)*cos(asp+45) + sin(15)*cos(grad); report shade.map = (shade-mapminimum(shade))/(mapmaximum(shade)-mapminimum(shade)); #### not used in lisem
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
108
Annex 7. Soil sampling site specific Cosine Similarity results
Site Properties PSD
Properties without
PSD
1 0.7 0.73 0.88 2 0.83 0.9 0.73 3 0.71 0.45 0.79 4 0.9 0.99 0.75 5 0.62 0.9 0.48 6 0.71 0.55 0.57 7 0.65 0.9 0.75 8 0.59 0.59 0.58 9 0.78 0.99 0.62 10 0.89 0.79 0.88 11 0.89 0.59 0.91 12 0.55 0.73 0.68 13 0.75 0.91 0.75 14 0.82 0.8 0.89 15 0.61 0.78 0.88 16 0.8 0.46 0.82 17 0.79 0.92 0.7 18 0.93 0.92 0.94 19 0.82 0.36 0.86 20 0.74 0.95 0.58 21 0.84 0.79 0.86 22 0.92 0.95 0.88 23 0.88 0.95 0.8 24 0.9 0.92 0.91 25 0.93 0.88 0.89 26 0.9 0.86 0.82 27 0.83 0.78 0.68 28 0.82 0.72 0.68 29 0.87 0.79 0.93 30 0.76 0.91 0.61 31 0.8 0.86 0.74 32 0.91 0.81 0.87 33 0.87 0.93 0.83 34 0.7 0.91 0.68 35 0.75 0.83 0.81 36 0.7 0.88 0.49 37 0.79 0.63 0.93 38 0.68 0.5 0.68 39 0.86 0.97 0.9 40 0.89 0.96 0.78 41 0.6 0.78 0.87 42 0.73 0.41 0.62 43 0.47 0.7 0.65 44 0.82 0.86 0.82 45 0.74 0.84 0.62 46 0.63 0.44 0.84 47 0.76 0.73 0.81 48 0.54 0.48 0.97
PSD= Particle Size Distribution, Properties included: Soil Organic Matter, Saturated Hydraulic Conductivity, Porosity, Bulk Density
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
109
Annex 8. Land cover map Laplae watershed (2005)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
110
Annex 9. Gumble plot of daily long-term precipitation measurements (1951-2018). Red circle represents the flash flood event in 2006.
y = 0.1819e0.0256x
R² = 0.9224
0
10
20
30
40
50
60
70
80
90
100
0 50 100 150 200 250 300
Ret
urn
Per
iod
(yr
.)
24h max rainfall (mm)
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
111
Annex 10. Simulated flood height measurements in comparison to measured flood height points
Point X-UTM Y-UTM Height Measured (m) Height Do (m) Height Dv (m) Height Dm (m) Height SRTM (m)
1 609784 1952841 0.704 0.0 0.0 0.0 0.0 2 609766 1952888 0.875 0.0 0.0 0.0 0.0 3 609828 1952893 1.165 1.8 0.1 2.3 0.3 4 609954 1952879 1.08 0.0 0.0 0.0 0.0 5 609819 1952776 0.505 0.0 0.0 0.0 0.1 6 610087 1952820 1.99 0.0 0.0 0.0 0.0 7 610203 1952754 2.35 0.0 1.0 0.0 0.0 8 610261 1952699 2.045 0.6 2.1 0.6 0.2 9 609924 1952528 0.545 0.0 0.0 0.0 0.2 10 609861 1952672 0.585 0.0 0.0 0.0 0.0 11 610251 1952598 1.99 0.0 0.2 0.0 0.0 12 610261 1952462 1.585 0.3 0.5 0.3 0.0 13 610114 1952239 0.6 0.3 0.3 0.3 0.1 14 610060 1952288 0.725 0.1 0.1 0.1 0.0 15 610009 1952344 0.635 0.5 1.3 0.5 0.2 16 609960 1952424 0.965 0.2 1.0 0.2 1.0 17 610278 1952376 1.52 0.0 0.0 0.0 0.0 18 610321 1952256 1.64 0.0 0.5 0.0 0.1 19 610638 1951968 1.91 0.3 0.0 0.5 0.1 20 610545 1951975 2.025 0.1 0.2 0.1 0.0 21 610465 1952010 1.7 0.0 0.0 0.0 0.0 22 610377 1952034 1.51 0.4 1.1 0.4 0.0 23 610334 1952056 1 0.0 0.5 0.0 0.1 24 610298 1952065 0.66 0.0 0.0 0.0 0.6 25 610283 1952101 0.86 0.2 0.2 0.2 1.6 26 610248 1952146 0.66 0.0 0.0 0.0 1.1 27 610163 1952196 0.67 0.5 0.5 0.5 0.0 28 610358 1952150 1.44 0.4 1.1 0.4 0.1 29 610365 1952075 1.375 0.4 1.1 0.4 0.0 30 610295 1952032 0.54 0.0 0.0 0.0 0.0 31 610738 1951944 1.85 0.3 2.2 0.5 0.0 32 610687 1951955 2.01 1.5 3.4 1.8 0.0 33 610272 1951929 0.525 0.0 0.0 0.0 0.1 34 610258 1951778 0.925 0.5 0.0 0.5 0.0 35 610536 1951933 1.575 0.1 0.1 0.1 0.1 36 610509 1951864 1.73 0.0 0.1 0.0 2.4 37 610487 1951768 1.305 0.3 0.4 0.3 0.0 38 610777 1951942 1.35 0.0 1.8 0.1 0.1 39 610255 1951668 0.88 1.8 1.4 1.8 3.1 40 610271 1951554 1.03 2.1 1.7 2.1 1.1 41 610486 1951667 1.46 0.9 1.2 0.9 1.0 42 610478 1951554 1.61 0.0 0.0 0.0 0.9 43 610277 1951446 1.1 0.8 0.4 0.8 0.0 44 610496 1951456 1.275 0.9 1.2 0.9 0.1 45 610519 1951379 1.255 0.0 0.0 0.0 1.3
Height measured = historical flood height, Height Do = simulated water height with original DEM, Height Dv = simulated water height with DEM without vegetation, Height Dm = simulated water height with manipulated DEM, Height SRTM = simulated water height with SRTM DEM
IMPACT ASSESSMENT OF SOIL INFORMATION AND LAND COVER CHANGE ON FLASH FLOOD MODELLING ON A WATERSHED SCALE ______________________________________________________________________________________________________________________________________________
112
Annex 11. Physical property layers of SoilGrids for soil layer 1 (5 cm)
Db = Bulk Density, CF = Coarse Fragments, Ks = Saturated Hydraulic Conductivity, SOC = Soil Organic Carbon